security assessment of cloud computing vendor offerings

34
Contract Number: N69250-08-D-0302 Task Order: 0007 Security Assessment of Cloud Computing Vendor Offerings Final Report Oct 10, 2009 Vassil Roussev, Ph.D. Golden G. Richard III, Ph.D. Daniel Bilar, Ph.D. Department of Computer Science University of New Orleans

Upload: others

Post on 03-Feb-2022

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Security Assessment of Cloud Computing Vendor Offerings

Contract Number: N69250-08-D-0302

Task Order: 0007

Security Assessment of

Cloud Computing Vendor Offerings

Final Report

Oct 10, 2009

Vassil Roussev, Ph.D.

Golden G. Richard III, Ph.D.

Daniel Bilar, Ph.D.

Department of Computer Science

University of New Orleans

Page 2: Security Assessment of Cloud Computing Vendor Offerings

1

Contents

Executive Summary ............................................................................................................. 2

Recommendations .............................................................................................................. 3

1. Introduction ................................................................................................................. 5

2. Background .................................................................................................................. 6

2.1. The Promise and Early Reality of the Cloud ..................................................................... 6

2.2. DoD Enterprises and the Cloud ........................................................................................ 9

2.3. Security Concerns on the Cloud ..................................................................................... 10

3. Representation of the Navy Data Center System Environment ................................ 11

4. Navy Security Concerns and Evaluation Criteria ....................................................... 12

5. Vendor Assessment Overview ................................................................................... 17

6. Vendor Assessment: Amazon Web Services (IaaS) ................................................... 17

6.1. Description ..................................................................................................................... 17

6.2. Security Assessment ...................................................................................................... 18

7. Vendor Assessment: Boomi Atmosphere (PaaS/SaaS).............................................. 20

7.1. Description ..................................................................................................................... 20

7.2. Security Assessment ...................................................................................................... 21

8. Vendor Assessment: force.com (PaaS) ...................................................................... 22

8.1. Description ..................................................................................................................... 22

8.2. Security Assessment ...................................................................................................... 23

9. Vendor Assessment: Pervasive Software (PaaS/SaaS) .............................................. 24

9.1. Description ..................................................................................................................... 24

9.2. Security Assessment ...................................................................................................... 25

9.3. Case Studies of Interest ................................................................................................. 26

Author Short Bio: Vassil Roussev ...................................................................................... 28

Author Short Bio: Golden G. Richard, III ........................................................................... 29

Author Short Bio: Daniel Bilar ........................................................................................... 30

List of Abbreviations ......................................................................................................... 31

References ........................................................................................................................ 32

Page 3: Security Assessment of Cloud Computing Vendor Offerings

2

Executive Summary

Cloud computing is a relatively new term that describes a style of computing in which a

service provider offers computational resources as a service over the Internet. The main

characteristic of cloud computing is the promise that service availability can be scaled up

and down according to demand with the customer paying based on actual resource usage

and service level arrangements.

Cloud computing differs from prior approaches to scalability in that it is based on a new

business proposition, which turns IT capital costs into operating costs. Under this model,

a service provider builds a public cloud infrastructure and offers its services to customers,

who only pay what they use in terms of compute hours, storage, and network capacity.

The expectation is that, by efficiently sharing the infrastructure, customers will see the

cost and complexity of their IT operations go down significantly.

From a security perspective, sharing the physical IT infrastructure with other tenants

creates new attack vectors that are not yet well understood. Virtualization is perceived as

providing a very high level of protection, yet it is too early to estimate its long-term

effectiveness: after all, it is a software layer subject to implementation faults. Cloud

services are in their very early stages of adoption, with small-to-medium enterprises as its

first customers. In all likelihood, no determined adversary has systematically tried to

penetrate the defense mechanisms of a cloud provider. Just as importantly, providers do

not appear ready to deliver the specific offerings that a DoD enterprise needs.

Over the medium and the long run, there is little doubt that DoD‟s major IT operations

will, ultimately, move to scalable cloud services that will increase capabilities and reduce

costs. However, current offerings from public cloud providers are not, in our view,

suitable for DoD enterprises and do not meet their security requirements.

Large civilian enterprises have very similar concerns to those of DoD ones and have

taken the path to develop private clouds, which use the same technology but the actual

operation stays inside the enterprise data center. In our view, this is the general direction

in which DoD entities should focus their efforts. It is difficult to conceive a scenario

under which it becomes acceptable for DoD data and computations to physically leave

the enterprise. Moreover, DoD operations have a large enough scale to reap the benefits

of cloud services by sharing the infrastructure within DoD.

It is important to recognize that a generic cloud computing platform, by itself, provides

limited benefits. These come from virtualization, which by consolidating existing

deployments reduces overall hardware requirements, simplifies IT management, and

provides greater flexibility to react to variations in demand.

The true cost savings come from well-known sources—eliminating redundant

applications, consolidating operations, and sharing across enterprises. In cloud terms, it is

multi-tenancy—the sharing of a platform by multiple tenants—that provides the greatest

efficiency gains. It is also important to recognize that efficient scaling of services requires

that applications be engineered for the cloud and it is likely that most legacy applications

would need to be reengineered.

Page 4: Security Assessment of Cloud Computing Vendor Offerings

3

Recommendations

The overall assessment of this report is that cloud computing technology has significant

potential to substantially benefit DoD enterprises. Specifically, it can facilitate the

consolidation and streamlining of IT operations and can provide operational efficiencies,

such as rapid automatic scaling of service capacity based on demand. The same

technology can simplify the deployment of surge capacity, as well as fail-over, and

disaster recovery capabilities. Properly planned and deployed, cloud computing services

could ultimately bring higher levels of security by simplifying and speeding up the

process of deploying security upgrades, and by reducing deployed configurations to a

small number of standardized ones.

It is important to both separate the potential of cloud computing from the reality of

current offerings, and to critically evaluate the implications of using the technology

within a DoD enterprise. Our own view (supported by influential IT industry research) is

that the current state of cloud computing offerings does not live up to level of hype

associated with them. This is not uncommon for new technologies that burst to the

forefront of the news cycle; realistically, however, it takes time for the technology to

mature and fulfill its promise. From a DoD perspective, the most important question to

ask is:

How does the business proposition of cloud computing translate into the DoD domain?

The vast majority of the benefits from using infrastructure-as-a-service (IaaS) offerings

can be had now by aggressively deploying virtualization in existing, in-house data

centers. Further efficiencies require higher level of sharing at the platform (platform-as-a-

service, PaaS) and application levels (software-as-a-service, SaaS). In other words, there

need to be multiple tenants that share these deployments and associated costs. Due to

trust issues, it is difficult to envision a scenario where DoD enterprises share deployments

with non-DoD entities in a public cloud environment.

Further, regulatory requirements with respect to trust, security, and verification have

hidden costs that are generally unaccounted for in the baseline (civilian) model. The

responsibility of assuring compliance cannot be outsourced and would likely be made

more difficult and costly.

Based on the above observations, we make the following general recommendations. A

DoD enterprise should:

Only consider deploying cloud services on facilities that it physically controls.

The general assumption is that the physical location of one‟s data in the cloud

should be irrelevant as users will have the same experience. This is patently not

true for DoD systems as the trustworthiness and security of any facility that hosts

data and computations must be assured at all times. Further yet, the supply chain

of the physical hardware must also be trustworthy to ensure that no breaches can

be initiated from it.

Consider vendors who offer technology to be deployed in a private cloud, rather

than public cloud services. Expanding on the previous point, it is inconceivable at

this point that DoD would relinquish physical control over sensitive data and

Page 5: Security Assessment of Cloud Computing Vendor Offerings

4

computations to a third party. Therefore, DoD should look to adopt the

technology of cloud computing but modify the business model for in-house use.

For example, IaaS technology is headed for commoditization with multiple

competing vendors (VMWare, Cisco, Oracle, IBM, HP, etc.) working on

technologies that automate the management of entire data centers. Such offerings

can be expected to mature within the next 1-2 years.

Approach cloud service deployment in a bottom-up fashion. The deployment of

cloud services is an evolutionary process, which ultimately requires re-

engineering of applications, as well as business practices. The low-hanging fruit is

data center virtualization, which decouples data and services from the physical

machines. This enables considerable consolidation of hardware and software

platforms and is the basic enabler of cloud mobility. Following that, it is the

automation of virtualized environment management that can bring the costs

further down. At the next level, shared platform deployments, such as database

engines and application servers, provider further efficiencies. Finally, the most

difficult part is the consolidation of applications and shared deployment of cloud

versions of these applications.

We fully expect that these themes are familiar to DoD IT managers and some

aspects of these are already implemented. This should not come as a surprise as

these are the true concerns of business IT efficiency and cloud computing does

not magically wave them away. However, cloud computing does provide extra

leverage in that, once a service is ready for the cloud, it can be deployed on a

wide scale at a marginal incremental cost.

Look for opportunities to develop and share private cloud services within DoD.

Sister DoD entities are natural candidates for sharing cloud service

deployments—most have common functions and compliance requirements, such

as human resources and payroll, that are prime candidates for multi-tenant

arrangements. Unlike the public case, the consolation of such operations can bring

the cost of compliance down as fewer systems would need to be certified.

Critically evaluate any deployments scenarios using Red Team exercises and

similar techniques. As discussed further in the report, most of the service

guarantees promised by vendors are based on the design of the systems and have

not been independently verified. This may be acceptable in the public domain as

the price of failure for many enterprises is high but rarely catastrophic for the

nation; however, DoD facilities must be held to a much higher standard. The only

realistic way of assessing how DoD cloud services would perform under stress, or

under a sustained attack by a determined adversary is to periodically simulate

those conditions and observe the system‟s behavior.

Page 6: Security Assessment of Cloud Computing Vendor Offerings

5

1. Introduction

Cloud computing is a relatively new term that describes a style of computing in which a

service provider offers computational resources as a service over the Internet. The main

characteristic of cloud computing is the promise that service availability can be scaled up

and down according to demand with the customer paying based on actual resource usage

and service level arrangements. Typically, the services are provided in a virtualized

environment with computation potentially migrating from machine to machine to allow

for optimal resource utilization. This process is completely transparent to the clients as

they see the same service regardless of the physical machine providing it. The name

„cloud computing‟ alludes to this concept as the Internet is often depicted as a cloud on

architectural diagrams, hence the computation happens „in the cloud‟.

Historically, the basic concept was pioneered by IBM in the 1960s under the name utility

computing, however, it has been no more than a decade since the cost and capabilities of

commodity hardware have made it possible for the idea to be realized on a massive scale.

In general, the service concept can be applied at three different levels and actual vendor

offering may be a combination of these:

Software as a Service (SaaS). Under the SaaS model, application software is

accessed and managed over the network and is usually hosted by a service

provider. Licensing is tied to the number of concurrent users, rather than physical

product copies. For example, Google provides all of its applications—Gmail,

Google Docs, Picasa, etc.—as services, and most of them are not even available

as standalone products.

Platform as a Service (PaaS). PaaS offers as a service a specific solution platform

on top of which developers can build applications. For example, in traditional

computing LAMP (Linux, Apache, MySQL, Perl/PHP/Python) is a popular choice

for developing Web applications. Example PaaS offerings include Google App

Engine, force.com, and Microsoft's Windows Azure Platform.

Infrastructure as a Service (IaaS). IaaS goes one level deeper and provides an

entire virtualized environment as service. The services can be provided at the

operating system level, or even the hardware level, where virtual compute,

storage, and communications resources can be rented on an on-demand basis.

Given the wide variety of offerings many of which are not even directly comparable, this

report provides a conceptual analysis of the field that relates to the Navy-specific

requirements. This drives the analysis of specific offerings and will be extend the useful

lifetime of this report by providing a framework for evaluating other offerings.

Page 7: Security Assessment of Cloud Computing Vendor Offerings

6

2. Background

"The security of these cloud-based infrastructure services is like Windows in 1999. It's being

widely used and nothing tremendously bad has happened yet. But it's just in early stages of

getting exposed to the Internet, and you know bad things are coming."

- John Pescatore, Gartner VP and security analyst, (quoted by Financial Times, Aug 3, 20091)

Before presenting the security implication of transitioning the IT infrastructure to the

cloud, it is important to understand the business case behind the trend and the

technological means by which it is being implemented. These have direct bearing on the

ultimate cost/benefit analysis with respect to security and it is important to understand

that some of the security assumptions behind the typical enterprise analysis may not hold

true for a DoD implementation. In turn, this may prevent at least some of the possible

efficiencies and may completely alter the outcome of the decision process.

2.1. The Promise and Early Reality of the Cloud

There are dozens of definitions of the Cloud and Vaquero et al. [22] recently completed

an in-depth survey on the topic. Based in part on those results, we offer one of the more

rigorous definitions:

“Clouds are a large pool of easily usable and accessible virtualized resources (such

as hardware, development platforms and/or services). These resources can be

dynamically reconfigured to adjust to a variable load (scale), allowing also for an

optimum resource utilization. This pool of resources is typically exploited by a pay-

per-use model in which guarantees are offered by the Infrastructure Provider by

means of customized service level agreements (SLA).”

The above statement can be broken into three basic criteria for a “true” cloud:

The management of the physical hardware is abstracted from the customer

(tenant); this goes beyond virtualization—the location of the data and the

computation is transparent to the tenant.

Infrastructure capacity is highly elastic allowing tenants to consume the almost

exact amount of resources demanded by users—the infrastructure automatically

scales up/down on a fine time scale.

Tenants incur no upfront capital expenditures—infrastructure costs are paid for

incrementally as an operating expense.

The main selling point of all cloud computing offerings is that they provide significant

reductions of the total cost of ownership, offer more flexibility, and have low entrance

costs for new enterprises. It is important to realize that the main source of efficiency in a

public cloud infrastructure comes from sharing the cost with other tenants.

Cloud platforms can provide savings at each layer of the hardware/software stack,

starting with the hardware, all the way up to the application layer:

Data center maintenance

1 http://www.ft.com/cms/s/0/5aa4f33e-7fc4-11de-85dc-00144feabdc0.html

Page 8: Security Assessment of Cloud Computing Vendor Offerings

7

Infrastructure/network maintenance

Infrastructure management and provisioning

Database management, provisioning, and maintenance

Middleware/application server management

Application management

Patch deployment and testing

Upgrade management

Typical data center arrangements in which racks of servers and storage are rented and

then managed by the tenant offer the lowest level of sharing and, therefore, the lowest

level of overall cost reduction. At the other end of the spectrum is a multi-tenant

arrangement in which multiple parties share the database, middleware, and application

platforms and outsource all maintenance to the provider. This arrangement promises the

greatest efficiency improvements. For example, Gartner Research reports [12] that its

clients are experiencing project savings of 25% to 40% by deploying customer

relationship management solutions via the SaaS model.

The real value of cloud computing kicks in when a full application platform is leveraged

to either purchase applications written to take advantage of a multi-tenant platform, or to

(re-)develop the applications on this new stack. For example, moving existing mail

servers from an in-house data center to a virtual machine somewhere in the cloud will not

yield any notable savings as the tenant would still be responsible for maintenance,

patching, and provisioning. The latter point is important—provisioning is done at peak

demand, the capability is utilized only a small fraction of the time, and the cost is borne

full time (According to McKinsey‟s survey [14], the average physical server is utilized at

about 10%.) In contrast, a true cloud solution, such as Google Mail, would relieve the

tenant from all of these concerns, potentially saving non-trivial amounts of physical and

human resources.

McKinsey‟s report offers a sobering view of current cloud deployments:

Current cloud offerings are most attractive for small and medium-sized enterprises;

adoption by larger enterprises (with the end goal of replacing the in-house data

centers) faces significant hurdles:

o Current cloud computing offerings are not cost-effective compared to large

enterprise data centers. (This conclusion is based on the assumption that existing

IT services would be replicated on the cloud and, therefore, most of the potential

efficiencies would not be realized.)

Figure 1 illustrates the point by considering Amazon‟s EC2 service for both

Windows and Linux. The overall conclusion is that most EC2 options are more

costly than TCO for a typical data center. It is possible to improve the cloud price

through pre-pay agreements for Linux system (but not for Windows). Further,

based on case studies, it is estimated that total cost of ownership (TCO) for 1

month of CPU processing is about $150 for an in-house data center and $366 on

Page 9: Security Assessment of Cloud Computing Vendor Offerings

8

EC2 for a comparable instance. Consequently, cloud costs need to go down

substantially before they can justify wholesale data center replacement.

Figure 1 EC2 monthly CPU equivalent price options [14]

o Security and reliability concerns will have to be mitigated and applications‟

architecture needs to be re-designed for the cloud. Section 2.3 discusses this point

in more detail.

o Business perceptions of increased IT flexibility and effectiveness will have to be

properly managed. Currently, cloud computing is relatively early in its technology

cycle with actual capabilities considerably lagging public perceptions. There is no

reason to doubt the momentum, or long-term viability of the technology itself as

all the major IT vendors—IBM, HP, Microsoft, Google, Cisco, VMWare, etc.—

are working towards providing integrated solutions. Yet, mature solutions for

larger enterprises appear a few years away.

o The IT organizations will have to adapt to function in a cloud-centric world

before all efficiencies can be realized.

Most of the gains in a private cloud deployment (using a cloud platform internally)

come from virtualizing server storage, network operations, and other critical building

blocks. In other words, by deploying virtualization aggressively, the enterprise can

achieve close to 90% of the utilization a cloud service provider (such as Google) can

achieve. Figure 2 illustrates the point.

The traditional IT stack consists of two layers: facilities—the actual (standardized)

data center facilities—and the IT platform—standardized servers, networking,

storage, and OS. Publicly-announced private clouds are essentially an aggressive

virtualization program on top of it—the virtualization layer contains hypervisors and

virtual machines in which OS instances execute.

Page 10: Security Assessment of Cloud Computing Vendor Offerings

9

Figure 2 Average server utilization rates [14]

2.2. DoD Enterprises and the Cloud

It should be clear that, as more and more of the functions are outsourced to the cloud

provider and are (implicitly) shared with other parties, a great deal of control over the

data and computation is handed over to the provider. For most private enterprises (the

most enthusiastic adopters so far) this may not present the same types of issues as for a

DoD entity:

Legal Requirements. It may not be permissible for sensitive government data

and processing to be entrusted to a cloud provider. Specifically, it is likely not

acceptable for DoD to relinquish control of their physical location and it is clear

that they would have to be guaranteed to stay on US territory. For example, some

current cloud computing providers, such as force.com, use backup data centers

outside of the U.S. as a disaster mitigation strategy. Further complicating the

picture is knowing where backup data is stored and this picture isn‟t clear—

force.com provides conflicting claims on this issue, stating both that backup tapes

never leave the data center [17] and that they are periodically moved offsite [18].

Google Apps terms of service state that it has the right to store and process

customer data in the United States "or in any other country in which Google or its

agents maintain facilities".

Compliance. DoD enterprises must meet specific standards in terms of trust and

security, starting with physical security and including security clearances, fine-

grained access control, and strict accounting. Such requirements are generally

lacking in the commercial space and, given the early stages of the technology

development cycle, any compliance claims on the part of providers need to be

scrutinized vigorously.

Legacy systems. Many DoD enterprises have to support and maintain numerous

legacy systems. Although the same is true for the commercial sector, it is not clear

that the DoD entity will have the resources and the freedom to replace those with

COTS solutions. Further, COTS solutions may not work as well (out of the box)

Page 11: Security Assessment of Cloud Computing Vendor Offerings

10

even for relatively standard functions like personnel and supply management,

simply because the government tends to do things differently.

Risk and Public Relations (PR) Risk. Objectively, the risk profile of DoD

agencies is considerably higher than that of private enterprises as national security

concerns are at stake. Accordingly, the public‟s tolerance for even minor breaches

and failures is almost non-existent. Therefore, decision makers must necessarily

put extra weight on the risk side of the equation and invest additional resources in

managing and mitigating those risks. Thus, a risk scenario that is statistically very

unlikely and can be ignored by private enterprises may require consideration (and

resources) in a DoD enterprise. With respect to IT, this would usually manifest

itself in the form of customized applications and procedures. Those are additional

costs that cloud computing is unlikely to remedy as they will have to replicated on

the cloud.

2.3. Security Concerns on the Cloud

Critical analysis reveals that cloud computing, both as a concept and as it is currently

implemented, brings a whole suite of security problems that stem primarily from the fact

that physical custody of data and computations is handed over to the provider.

Physical security. The first problem is that we have a completely new physical security

perimeter—for most enterprises this may not be noticeable but for a DoD entity

considerations could be different. First, is the location where the data is housed

physically as secure as the in-house option? If the computation is allowed to „float‟ in the

cloud, are all the possible locations sufficiently secure?

Confidentiality. Traditionally, operating systems are engineered such that a super-user

has unlimited access to all resources. This approach was recognized as a problem in

database systems so trusted DBMS were developed so that the administrator could

manipulate the structure of the database but not have (by default) access to the records.

The super-user problem escalates in the cloud environment—now the hypervisor

administrator has control over all OS instances under his command. Further, if the

computation migrates, another administrator has access to it, and so on. The concept of

trusted cloud computing has been recently put forward as a research idea [19] but, to the

best of our knowledge, no technical solutions are yet deployed.

Another concern is the physical location of the data—if it is migrated and replicated

numerous times, what are the guarantees that no traces will be left behind? Along the

same lines, OS instances are routinely cloned and suspended and the memory images

contain sensitive data—keys, unique identifiers, password hashes, etc.

New attack scenarios. In the cloud, we have new neighbors that we know nothing

about—normally, if they are malicious, they would have to work hard to breach the

security perimeter before they can launch an attack. In the cloud, they are already sharing

the infrastructure and can exploit security flaws virtual machine hypervisors, virtual

machines, or in third-party applications for privilege escalation and attack from the cloud.

In public cloud computing services, such as Amazon‟s, the barriers to entry are very low.

The services are highly automated and all that is needed is a credit card to open an

account and start probing the infrastructure from within.

Page 12: Security Assessment of Cloud Computing Vendor Offerings

11

Limits on security measures. Not having access to the physical infrastructure implies that

traditional measures that rely on physical access can no longer be deployed. Security

appliances, such as firewalls, IDS/IPS, spam filters, data leak monitors, can no longer be

used. Even for software-based protection mechanisms there are no solutions that allow

the tenant to control policy decisions. This has been recognized as a problem and

VMWare has recently introduced the VMSafe interface, which allows third-party

modules to control security policy decisions. Even so, tenants need also to be concerned

about the internal security procedures of the provider.

Testing and compliance. Part of ensuring the security of a system is to continuously test it

and, in some cases, such testing is part of the compliance requirement. By default,

providers prohibit malicious traffic and possibly even vulnerability scans [2], although it

may be possible to reach an arrangement as part of the SLA.

Incident response. Early detection and response is one of the critical components of

robust security and the cloud makes those tasks more challenging. To perform an

investigation, the administrator will have to rely on logs and other information from the

provider. It is essential that a) the provider collects the necessary information; and b) the

SLA provides for that information be accessible in a timely manner.

Technical challenges. OS security is built for the physical hardware and in its transition

to the virtual environment there are several challenges that are yet to be addressed:

Natural sources of randomness. Cryptographic implementations rely on random

numbers and it is the job of the OS to collect random data from non-

deterministic events. In a virtualized environment, it is possible that cloned

instances would behave in a predictable-enough manner to execute attacks [20]

Trusted platform modules (TPM). Since the physical hardware is shared among

several instances the TPM-supported remote attestation process would have to

be redesigned.

3. Representation of the Navy Data Center System Environment

Currently, SPAWAR New Orleans supports over two dozen applications running in a

virtualized data center environment. At any given time, up to 300 VM instances are in

execution. Over time, it is expected that both the number of applications and the

workload will continue to grow steadily and increase demand on the operation. Part of

the mission is to support incident management, such as hurricane response, which

requires that redundant and reliable capacity be available. It is self-evident that, for most

of the applications, it is critical that they be available and functioning correctly under any

circumstances.

The baseline platform configuration consists of commodity Sun Microsystems and Dell

hardware, and Sun Solaris and Microsoft Windows operating systems. The virtualization

layer is implemented using VMWare products. Data storage is organized using a

centralized Storage Area Network, with local disk boot configuration and SAN-based

application installations.

Incident Response

Page 13: Security Assessment of Cloud Computing Vendor Offerings

12

Incident response systems support communication, coordination, and decision making

under emergency conditions, such as natural disasters. Specifically, Navy provides a

messaging platform, which allows for information to be broadly disseminated among first

responders. It also provides tracking of assets with information provided by transponders

and provides up-to-date information to decision makers through appropriate visualization.

Overall, most of the information exchanged via incident response application is of a

sensitive nature and, as such, must be closely monitored and regulated. Specific concerns

include the location of assets, personally-identifiable information, as well as operational

security. The latter presents a difficult challenge as it often has to do with the

accumulation of individual data points over time that can, together, paint a bigger picture

of the enterprise. This data may include patterns of response, organizational structure,

and availability of assets that would normally be kept secret.

The threat to operational security in this context is well illustrated by recent research on

hidden databases on the Internet [13]. A hidden database is one that is not generally

available but does provide a query interface, such as a Web form, that can return a limited

sample. An example would be any reservation system which provides information on

availability for specific parameters. Researchers have shown that it is quite feasible to

devise automated queries that can exploit the limited interface to obtain representative

samples of the underlying database and to accurately estimate the overall content of the

hidden database. The solution to such concerns involves careful design of the application

interface so that returned data does not permit an adversary to collect representative

samples. Absent specifics of the Navy application pool, this concern is not discussed

further in this report.

Logistics

Navy data centers provide logistic systems for Navy operations by tracking assets,

equipment, and maintenance schedules. The execution of such applications has concerns

similar to those discussed in the previous section, especially operational security.

Human Resources

Human resources applications provide the typical HR functions of a large enterprise as

applied to the specific needs of the Navy. In addition to hiring and discharging of service

members, the system keeps track of skill sets and pay certifications. An additional

concern is the non-disclosure of personally-identifiable information to third parties. This

includes restricting the release of information that can indirectly lead to such disclosure.

As with other operational security concerns, the counter-measures are application-

specific and are not discussed further.

4. Navy Security Concerns and Evaluation Criteria

In general terms, the Navy‟s IT operations have several major security concerns, some of

which are specific to its operation as a DoD enterprise. There are a number of factors that

a repeatable decision making process should incorporate when assessing cloud computing

vendor offerings. As an unclassified report, there are limits on the amount of specific

information that we could obtain and provide. Therefore, the outlined criteria should be

Page 14: Security Assessment of Cloud Computing Vendor Offerings

13

treated as a high-level roadmap and the specific details need to be fleshed out when an

actual decision point is at hand

Release-code Auditing

One of the main concerns is release-code auditing which requires that all source code be

audited before being put into production. This is, effectively, a major restriction on the

kind of services that can be provided by commercial vendors on general-purpose cloud

platforms.

It is likely economically and logistically infeasible for vendors to undergo a certification

process for all the code that could provide services to the Navy. For example, it would

not be feasible to use an existing scalable email platform, such as the one provided by

Google to improve the scalability of Navy‟s operations.

Compliance with DoD Regulations

Another hurdle in the adoption of COTS cloud services is the need to ensure that all

services comply with internal IT requirements. Due to the sensitive nature of this

information, no specific issues are discussed in this report. It is worth noting, however,

that even if the problem can be resolved administratively and the vendor can demonstrate

compliance, the Navy would still have to dedicate resources to monitor this compliance

on a continuous basis.

The Common Criteria for Information Technology Security Evaluation (CC) is an

international standard (ISO/IEC 15408) for computer security certification. CC is a

framework in which computer system users can specify their security requirements,

vendors can then implement and/or make claims about the security attributes of their

products, and testing laboratories can evaluate the products to determine if they actually

meet the claims. DoD security requirements can be expressed, in part, by referring to

specific protection profiles and can help determine minimum qualifications. For example,

VMWare‟s ESX Server 3.0.2 and VirtualCenter 2.0.2 have earned EAL4+ recognition,

whereas Xen, the virtualization solution used by Amazon, has not been certified.

Overall, the CC certification is a long and tedious process and speaks more directly to the

commitment of the vendor rather than the quality of the product. As a 2006 GAO report

[23] points out in its findings, there is “a lack of performance measures and difficulty in

documenting the effectiveness of the NIAP process”. Another side effect of certification

is that it takes up to 24 months for a product to go through an EAL4 certification process

[23] (p.8), which limits the availability of products with up-to-date features.

Given the fast pace of development in cloud computing products, basing a decision solely

on certification requirements may severely constrain DoD‟s choice. One possible

approach is to formulate narrower, agency-specific certification requirements to speed up

the certification. It is also important to recognize that the common EAL4 certification

does not cover some of the more sophisticated attack patterns that are relevant to cloud

computing, such as side channel attacks.

Another existing regulation that could provide a ready reference point for evaluation is

The Health Insurance Portability and Accountability Act of 1996 (HIPAA). HIPAA is a

set of established federal standards, implemented through a combination of

administrative, physical and technical safeguards, intended to ensure the security and

Page 15: Security Assessment of Cloud Computing Vendor Offerings

14

privacy of protected health information. These standards affect the use and disclosure of

protected health information by certain covered entities (such as healthcare providers

engaged in electronic transactions, health plans and healthcare clearinghouses) and their

business associates.

In technical terms, HIPAA requires the encryption of sensitive information „in-flight‟ and

„at-rest‟, dictates basic administrative and technical procedures for setting and enforcing

access control policies; and requires in-depth auditing capabilities, data back-up

procedures and disaster recovery mechanisms. These requirements are set based on

established best security practices and are common to all applications dealing with

sensitive information. From a DoD perspective, these are minimal standards and

additional safeguards are likely necessary for most applications. In that respect, HIPAA

can be considered a minimal qualification requirement.

Overall, the emergence of any major standards targeted at cloud computing should be

taken as a cue that cloud computing is reaching a certain level of maturity, as standards

development tends to follow rather than lead. However, the emergence of standards

cannot, by itself, be relied upon as a timing mechanism to identify optimal points for

adopting new technologies. Standards processes have mixed record on their timeliness (at

best) and tend to follow rather than lead. In our view, given the current pace of

development by major IT companies, it is likely that, within the next two years, mature

offerings will start to differentiate themselves from the rest.

Limits on Physical Security

Currently, the security of the services is provided, in part, through the use of appliances,

such as firewalls and intrusion detection systems. These devices are used to

compartmentalize the execution of different applications and isolate faults and breaches.

Having a separate and specialized device creates an independent layer of protection that

considerably raises the bar for attackers—a wholesale breach requires that multiple layers

be breached.

In a cloud deployment, there is no direct analogy to this practice although the industry

has recognized the problem and is moving toward offering solutions. Specifically,

virtualization vendors are beginning to offer an interface (API) through which third-party

virtual appliances could be plugged-in to enforce security policies and monitor traffic and

execution environments. The conceptual problem is that such components are dependent

on the security of the hypervisor—if it is breached there is the opportunity for attackers to

circumvent the virtual security appliances. Eventually, solutions to this problem will

likely be found—CPU vendors, such as Intel, are actively developing hardware support

for virtualization. Yet, at this relatively early stage of development, there is no proof that

virtual appliances offer the same level of protection as physical ones.

Underlying Hardware Security and Trust

Since cloud infrastructure is likely to be remote, it is impractical/impossible for it to be

scrutinized by the Navy to comply with its heightened level of security concerns and the

issue of hardware trojans must be addressed. With the advent of global markets,

vertically integrated chip manufacturing has given way to a cost-efficient, multi-stage

Page 16: Security Assessment of Cloud Computing Vendor Offerings

15

supply chain whose individual stages may or may not pass through environments friendly

to US interests (Figure 3)

Figure 3: IC manufacturing supply chain [9]

A DoD report crystallizes the issues thusly [9]:

" Trustworthiness of custom and commercial systems that support military

operations [..] has been jeopardized. Trustworthiness includes confidence that

classified or mission critical information contained in chip designs is not

compromised, reliability is not degraded or untended design elements inserted in

chips as a result of design or fabrication in conditions open to adversary agents.

Trust cannot be added to integrated circuits after fabrication; electrical testing and

reverse engineering cannot be relied upon to detect undesired alterations in

military integrated circuits. The shift from United States to foreign IC

manufacture endangers the security of classified information embedded in chip

designs; additionally, it opens the possibility that 'Trojan horses' and other

unauthorized design inclusions may appear in unclassified integrated circuits used

in military applications. More subtle shifts in process parameters or layout line

spacing can drastically shorten the lives of components."

If for conventional IT delivery platforms which are locally accessible only design

specifications and the testing stages can be controlled, the situation is exacerbated for

remote cloud computing platforms: A very real possibility of hardware-based malicious

code, hiding in underlying integrated circuits, exists. Potentially serious ramifications of

hardware-based subversion include unmitigatable cross-tenant data leakage to VM

identification and denial of service attacks [16].

We stress that this type of surreptitious hardware subversion is not within the ability of

AV software of cloud vendors to deal with; it is an ongoing, open research problem to

detect such malicious code in hardware at all. It is exacerbated, however, by the lack of

infrastructure control inherent in using public cloud offerings, as well as the motivation

of the cloud infrastructure owner to purchase and deploy COTS available.

Client-side Vulnerabilities Mitigation

Cloud clients (in particular web browsers) incorporate more functionality than the mere

display of text and images, including rich dynamic content comprising media playback

Page 17: Security Assessment of Cloud Computing Vendor Offerings

16

and interactive page elements such as drop-down menus and image roll-overs. These

features includes extensions such as the Javascript programming language, as well as

additional features for the client such as application plugins (Acrobat Reader,

QuickTime, Flash, Real, and Windows Media Player), and Microsoft-specific

enhancements such as Browser Helper Objects and ActiveX (Microsoft Windows‟s

interactive execution framework). Invariably, these extensions have shown

implementation flaws that can be maliciously exploited as security vulnerabilities to gain

unauthorized access to obtain sensitive information.

In August 2009, Sensepost demonstrated a proof-of-concept password brute-forcing with

password reset links—most if not all cloud apps use some password recovery (email- or

secret questions-based), several ways to steal cloud resources: Amazon cloud instance of

Windows license stealing, paid application theft (via DevPay), “cloud DoS” with

exponential virus-like growth of EC2-hosted VM instances [4].

When a user accesses a remote site with a client (say a browser), he types in a URL, and

initiates the connection. Once connected, a relationship of trust is established: the user

and the website (the user initiated the connection, and now trusts the page and content

display) and conversely, the site and the user (in executing actions from the user‟s

browser). It is this trust, together with the various features incorporated into rich clients

that attackers could subvert through what is called Cross-Site Scripting (XSS) and Cross-

Site Request Forgery (CSRF) attacks.

Cross-site scripting attacks mostly use legitimate web sites as a conduit, where web sites

allow other (malicious) users to upload or post links on to the web site. It must be

emphasized that from the point of view of the clients, neither HTTPS (the encrypted

channel with the little lock in the browser that denotes „safety‟) nor logins protect against

XSS or CSRF attacks. In addition, unlike XSS attacks which necessitate user action by

clicking on a link, CSFR attacks can also be executed without the user‟s involvement,

since they exploit explicit software vulnerabilities (i. e. predictable invocation structures)

on the cloud platform. As such, the onus to prevent CSFR attacks falls squarely on the

cloud vendor application developers. Some login and cryptographic token approaches, if

conscientiously designed to prevent CSFR attacks, can be of help [5].

Availability, Scalability, and Resistance to Denial of Service Attacks

Practically all service providers claim up time well above 99%, scaling on demand, and

robust network security. By and large, such claims are based on a few architectural

features (redundant power supply, multiple network connections, presence of network

firewalls) and should be scrutinized with a healthy dose of skepticism. Few (if any) of the

providers have performed large-scale tests to quantify how their services really behave

under stress. Customers should demand concrete proof that the provider can fulfill the

promises in the service level agreement (SLA).

In our view, the only way to adequately assess the performance-related vulnerabilities is

to perform a Red Team exercise. Such an exercise, unlike the SLA, would be able to

answer some very specific questions:

How many simultaneous infrastructure failures are necessary to take the system

down?

Page 18: Security Assessment of Cloud Computing Vendor Offerings

17

How quickly can it recover?

How does the system perform under excessive load—does performance degrade

gracefully, or does it collapse?

How well does the system really scale? In other words, as the load grows, how

fast do resource requirements grow?

What happens if other tenants misbehave—does that affect performance?

How well can the service resist a massive brute-force DoD attack? The latter is

not just a function of the deployed hardware and software but also includes the

adequacy and the training of administrative staff and their ability to communicate

and respond.

Short of a complete Red Team simulation, an experienced technical team should examine

in detail the architecture of the system and try to answer as many of the above questions

as possible. This will not give the same level of assurance but should be minimum

assessment performed on an offering selected for potential acquisition.

5. Vendor Assessment Overview

The presented assessments should be viewed as a small but typical sample of the types of

services offered on today‟s market. The research effort was hampered by a relative dearth

of specific information when it comes to the security of offered services. Although our

sample of surveyed products was by no means exhaustive, we found it difficult to extract

information from vendor representative beyond what is already publicly available. Some

completely ignored our requests for information and were not included here.

6. Vendor Assessment: Amazon Web Services (IaaS)

6.1. Description

There are still differing opinions on what exactly constitutes cloud computing, yet there

appears to be a consensus that, whatever the definition might be, Amazon‟s Web Services

(AWS) are a prime example. Frequently, IaaS are also referred to as utility computing.

Broadly, AWS consists of several basic services:

Amazon Elastic Computing Cloud (EC2) is a service that permits a customer to

create a virtual OS disk image (in a proprietary Amazon format) that will be

executed by a web service running on Amazon‟s hosted web infrastructure.

Amazon Elastic Block Store (EBS) provides block level storage volumes for use

with EC2 instances. EBS volumes are off-instance storage that persists

independently from the life of an instance. Volumes can range from 1 GB to 1 TB

that can be mounted as devices by EC2 instances. Multiple volumes can be

mounted to the same instance.

Amazon SimpleDB is a web service providing the core database functions of data

indexing and querying. The database is not a traditional relational database

although it provides an SQL-like query interface.

Amazon Simple Storage Servic (S3) is a service that essentially provides an

Internet-accessible remote network share.

Page 19: Security Assessment of Cloud Computing Vendor Offerings

18

Amazon Simple Queue Service (SQS) is a messaging service which offers a

reliable, scalable, hosted queue for storing messages as they travel between

computers. The main purpose is to automate workflow processes and provides

means to integrate it with EC2 and other AWS.

Amazon Elastic MapReduce is a web service that emulates the MapReduce

computional model adopted by Google. It utilizes a hosted Hadoop2 framework

running on the EC2 infrastructure. Hadoop is the open source implementation of

the ideas behind Google‟s MapReduce and is supported by major companies, such

as IBM and Yahoo!.

Amazon CloudFront is a recent addition which automates the process of creating a

content distribution network.

Amazon Virtual Private Cloud (VPC) enables enterprises to connect their existing

infrastructure to a set of isolated AWS compute resources via a Virtual Private

Network (VPN) connection, and to extend their existing management capabilities

such as security services, firewalls, and intrusion detection systems to include

their AWS resources.

It is fair to say that Amazon‟s services interfaces are emerging as one of the early de

facto technical standards of cloud computing. One recent development that can further

accelerate this trend is the development of Eucalyptus (Elastic Utility Computing

Architecture Linking Your Programs To Useful Systems). It is an open-source software

infrastructure for implementing cloud computing on clusters. The current interface to

Eucalyptus is compatible with Amazon's EC2, S3, and EBS interfaces, but the

infrastructure is designed to support multiple client-side interfaces. Eucalyptus is

implemented using commonly available Linux tools and basic Web-service technologies

and is becoming a standard component of the Ubuntu Linux distribution.

6.2. Security Assessment

The primary source of information for this assessment is the AWS security whitepaper

[3], which focuses on three traditional security dimensions: confidentiality, integrity, and

availability. It does not cover all aspects of interest but does provide a good sampling of

the overall state of affairs.

Certifications and Accreditations

Amazon is cognizant of its customers‟ need to meet certification requirement, however,

actual certification efforts appear to be at an early stage:

“AWS is working with a public accounting firm to ensure continued Sarbanes Oxley

(SOX) compliance and attain certifications such as recurring Statement on Auditing

Standards No. 70: Service Organizations, Type II (SAS70 Type II) certification.”

Separately, Amazon provides a short white paper on building HIPAA-compliant

application [1]. From the content, it becomes clear that virtually all responsibility for

comply with the regulations falls on the customer.

AWS provides key-based authentication to access their virtual servers. Amazon EC2

creates a 2048 bit RSA key pair, with private and public keys and a unique identifier for

2 http://hadoop.apache.org/

Page 20: Security Assessment of Cloud Computing Vendor Offerings

19

each key pair to facilitate secure access. The default setup for Amazon‟s EC2‟s firewall is

deny-all mode which automatically denies all inbound traffic unless the customer

explicitly opens an EC2 port. Administrators can create multiple security groups in order

to enforce different ingress policies as needed and can control each security group with a

PEM-encoded X.509 certificate and restrict traffic to each EC2 instance by protocol,

service port, or source IP address.

Physical Security

Physical security measures appear consistent with best industry practices but clearly do

not provide the same level of physical protection as an in-house DoD facility:

“AWS data centers are housed in nondescript facilities, and critical facilities have

extensive setback and and military grade perimeter control berms as well as other

natural boundary protection. Physical access is strictly controlled both at the

perimeter and at building ingress points by professional security staff utilizing video

surveillance, state of the art intrusion detection systems, and other electronic means.

Authorized staff must pass two-factor authentication no fewer than three times to

access All visitors and contractors are required to present identification and are signed

in and continually escorted by authorized staff.”

Backups

Backup policies are not described in any level of details and the provided brief

description is contradictory. It is only clear that some of the data is replicated at multiple

physical locations, while the remaining part is customer responsibility.

Amazon Elastic Compute Cloud (EC2) Security

EC2 security is the most descriptive part of the white paper and provides the greatest

amount of useful information although many details are obscured.

Amazon‟s virtualization layer utilizes a “highly customized version” of the Xen

hypervisor (http://xen.org). Xen is based on the para-virtualization model in which the

hypervisor runs inside a host operating system (OS) and provides a virtual hardware

interface to guest OS instances. Thus, all privileged access is mediated and controlled by

the hypervisor. The firewall is part of the hypervisor layer and mediates all traffic

between the network interface and the guest OS. All guest instances are isolated from

each other.

Virtual (guest) OS instances are built and are completely controlled by the customer as

physical machines would be. Customers have root access and all administrative control

over additional accounts, services, and applications. AWS administrators do not have

access to customer instances, and cannot log into the guest OS.

The firewall supports groups, thereby permitting different classes of instances to have

different rules. For example, in the case of a traditional three-tiered web application, web

servers would have ports 80/443 (http/https); application servers would have an

application-specific port open only to the web server group; database servers would have

port 3306 (MySQL) open only to the application server group. All three groups would

permit administrative access on port 22 (SSH), but only from the customer‟s corporate

Page 21: Security Assessment of Cloud Computing Vendor Offerings

20

network. The firewall is cannot be controlled not by the host/instance itself—it requires

the customer's X.509 certificate and key to authorize changes.

AWS provides an API that allows automated management of virtual machine instances.

Calls to launch and terminate instances, change firewall parameters, and perform other

functions must be signed by an X.509 certificate or the customer‟s Amazon Secret

Access Key and can be encrypted in transit with SSL to maintain confidentiality.

Customers have no access to physical storage devices and the disk virtualization layer

automatically wipes no longer in use to prevent data leaks. It is recommended that tenants

use encrypted file systems on top of the provided virtual block device to maintain

confidentiality.

Network security mechanisms address/mitigate the most common attack vectors. DDoS

are mitigated using known techniques such as syn cookies and connection limiting. In

addition, Amazon maintains some additional spare network capacity. VM instances

cannot spoof their own IP/MAC addresses as the virtualization layer enforces correctness.

Port scanning is generally considered ineffective as almost all ports are closed by

defaults. Packet sniffing is not possible as the virtualization layer will effectively prohibit

setting the virtual NIC from being put into promiscuous mode—no traffic that is not

addressed to the instance will be delivered. Man-in-the-middle attacks are prevented

through the use of SSL-encrypted communication.

Amazon S3/SimpleDB Security

Storage security, as represented by the S3 and SimpleDB services, has received relatively

light treatment. Data at rest is not automatically encrypted by the infrastructure and it is

the application‟s responsibility to do that. One drawback is that, if application data is

stored encrypted on SimpleDB, the query interface is effectively disabled.

The S3 APIs provide both bucket- and object-level access controls, with defaults that

only permit authenticated access by the bucket and/or object creator. Write and Delete

permission is controlled by an Access Control List (ACL) associated with the bucket.

Permission to modify the bucket ACLs is itself controlled by an ACL, and it defaults to

creator-only access. Therefore, the customer maintains full control over who has access

to their data. Amazon S3 access can be granted based on AWS Account ID, DevPay

Product ID, or open to everyone.

When an object is deleted from Amazon S3, removal of the mapping from the public

name to the object starts immediately, and is generally processed across the distributed

system within several seconds. Once the mapping is removed, there is no external access

to the deleted object. That storage area is then made available only for write operations

and the data is eventually overwritten by newly stored data.

7. Vendor Assessment: Boomi Atmosphere (PaaS/SaaS)

7.1. Description

Boomi AtomSphere offers an on-demand integration platform for any combination of

Software-as-a-Service, Platform-as-a-Service, Infrastructure as a Services, and on-

premise applications. Their main selling point is leveraging existing applications by

Page 22: Security Assessment of Cloud Computing Vendor Offerings

21

providing connectors for integration of SaaS offerings with on premise, back office

applications. This is a scenario that may be attractive for Navy purposes.

The integration processes are implemented through their proprietary, patent-pending

Atom, a dynamic runtime engine that can be deployed remotely or on premises. These

Atoms capture the components of end-to-end integration processes, including

transformation and business rules, processing logic and connectors.

They can be hosted by Boomi or other cloud vendors for SaaS-to-SaaS integration or

downloaded locally for SaaS to on-premise integrations. On-premise applications are

typically firewalled, with no direct access via the Internet, and no access even via a DMZ.

To handle this requirement, the Boomi Atom can be deployed on-premise to directly

connect the on-premise applications with one or more SaaS/Cloud applications. Changes

to firewalls (such as opening an inbound port) are not required, and the Atom supports a

full bi-directional movement of data between the applications being integrated. Deployed

locally, no data enters Boomi's data center at any point.

Should SaaS to SaaS integration be required and such applications accessed via a secure

Internet connection, Atoms can be hosted in Boomi's cloud, with Boomi managing the

uptime of the Atom. Customer data is isolated from other tenants in Boomi's platform

(though standard caveats apply, as mentioned above )

Finally, it is possible deploy Atoms into any cloud infrastructure that supports Java, such

as Amazon, Rackspace, and OpSource, offering direct non-Boomy connectivity between

between applications. Also in this deployment style, no customer data enter Boomi's data

center.

7.2. Security Assessment

Boomi has deployed multiple controls at the infrastructure, platform, application, and

data level, thus acknowledging multidimensional security aspects of their product.

Infrastructure Security

The Boomi infrastructure meets the AICPA SAS70 Type II (and Level l PCI DSS) audit

requirements, which is the most widely recognized regulatory compliance mandate issued

by the American Institute of Certifed Public Accountants. Its primary scope is the

inquiry, examination, and testing of service organization‟s control environment. Data

centers, managed-service providers, and SaaS vendors represent such service

organizations [15].

Boomi's controls include best-of-breed routers, firewalls and intrusion detection systems,

DDoS protection bolstered by redundant IP connections to world class carriers terminated

on a carrier grade network. Physical power continuity is provided by redundant UPS

power, diesel generator backups, as well as HVAC facilities. In addition, they have

instated multipoint monitoring of key metrics alerts for both mission critical and ongoing

maintenance issues.

Platform and Application Security

As noted, Atoms can reside locally or be hosted in Boomi's data center. An Atom can

communicate information to the Boomi data center in two modes, continuous automatic

Page 23: Security Assessment of Cloud Computing Vendor Offerings

22

and user-initiated communications. During ongoing communications, Atoms merely send

operational information such as online uptime status, tracking information cataloging

process executions, configuration updates, and code update checks to Boomi. User-

initiated communication is undertaken only upon the request of an authorized user. The

information sent includes logging information about specific integration processes, error,

failure and diagnostic messages, and retrieving schemata for the design of new

integration processes.

No inbound firewall ports need to be opened in order for an Atom to communicate with

Boomi's data center. Traffic is protected by standard 128 bit SSL encryption. Any

credential password needed for application integration (like an database password) is

encrypted by x509 private/public key pairs and stored for the account. When an Atom is

deployed, the encrypted password is passed along and the credentials supplied unlock the

password at runtime.

The AtomSphere platform (used to build, deploy and manage the Atoms, regardless of

deployment style) is accessed via a standard web browser. Boomi uses the OWASP Top

Ten list to address the most critical client and server side web security application flaws.

The U.S.Defense Information Systems Agency recommends OWASP Top Ten as key

best practices to be used as part of the DOD Information Technology Security

Certification and Accreditation Process (DITSCAP, now DIACAP) [10].

Cross-site Scripting (CSS) and Cross-Site Forgery Request (CSFR) mentioned earlier are

also listed in the Top Ten. Boomi's controls to prevent CSS relies on proper XML

encoding through an authenticated AWS REST-based API when data is delivered to the

client. Timestamps, as well as the aforementioned AWS authentication are used to

mitigate (though not eliminate) CSRF attacks [6].

Client Data Security

Boomi stresses that its AtomSphere platform does not by default retrieve, access or store

client data. It merely supports necessary data mapping rules to facilitate integration

without saving data at Boomi's location, unless specifically configured to do so. Hence,

data flowing through Atoms residing locally do not touch the Boomi data center: It is

transported directly to either the SaaS or a local application through an Atom component

(a connector) configured to user-specified security requirements. Should the client prefer

a zero-footprint Boomi data center hosted Atom deployment, the data center

infrastracture controls are used to safeguard the integrity, confidentiality and availability

of those Atoms.

8. Vendor Assessment: force.com (PaaS)

8.1. Description

force.com offers a Platform-as-a-Service cloud architecture to clients, designed to support

turn-key, Internet-scale applications. Their primary selling point is a track record of high

system uptime, with an Internet-facing mechanism for tracking reliability, and the ability

Page 24: Security Assessment of Cloud Computing Vendor Offerings

23

to declaratively develop certain classes of web-based applications, with little need to

write code.

Client applications to be executed on the force.com cloud are stored as metadata, which is

interpreted and transformed into objects executed by the force.com multitenant runtime

engine. Applications for the force.com architecture can be developed declaratively using

a native application framework, via a Java-like programming language called Apex, or

via exposed APIs that allow applications to be developed in C#, Java, and C++. The

APIs support integration with other environments, e.g. to allow data to be accessed from

sources external to the force.com infrastructure. Applications that use the API do not

execute within the force.com cloud—they must be hosted elsewhere.

force.com imposes a strict application testing regiment on new applications before they

are deployed, to ensure that new applications do not seriously impact the performance of

existing applications running in the force.com cloud. An extensive set of resource limits

is also imposed, to prevent applications from monopolizing CPU resources and memory.

Operations that violate these resource limits result in runtime exceptions in the

application.

8.2. Security Assessment

force.com deploys a number of mechanisms for increasing the security of applications

and associated data. These are described in the following sections.

Infrastructure Security

The force.com instrastructure‟s network is secured via external firewalls that block

unused protocols and the deployment of internal intrusion detection sensors on all

network segments. All communication with force.com is encrypted, via SSL/TLS. Third

party certification is regularly performed to assess network security. Power and HVAC

for datacenters is fully redundant, with multiple UPS‟s, power distribution units, diesel

generators, and cooling systems. External network connectivity is provided via fiber

enclosed in concrete vaults. A number of physical security measures are deployed at

force.com datacenters, including 24 hour manned security, biometric scanning for access

to computer systems, full video surveillance, and bullet-proof, concrete-walled rooms.

Computers hosting cloud-based applications are enclosed in steel cages with

authentication control for physical access. On the other hand, a successful phishing

attack has been mounted against force.com employees, resulting in the leakage of a large

amount of customer contact data [11].

Platform and Application Security

Native force.com applications are stored as metadata and executed by a runtime engine.

The database-oriented nature of the force.com APIs and the lack of low-level APIs for

applications executing within the cloud severely limits the possibility of; force.com

applications do not execute independently of the runtime engine (which has extensive

auditing and resource monitoring checks) and applications developed using force.com

APIs do not execute within the force.com cloud—they merely have access to data in the

cloud. On the other hand, if an attack vector against the multitenant runtime engine itself

were developed, then it appears that data and applications belonging to other

Page 25: Security Assessment of Cloud Computing Vendor Offerings

24

organizations could be manipulated, since data is comingled. No attack vectors of this

kind have been reported and the feasibility of developing such attacks is unknown.

Client Data Security

Measures to ensure client data security are vague—available force.com literature simply

states that “salesforce.com protects customer data by ensuring that only authorized users

can access it” and that “All data is encrypted in transfer.” One mechanism that might

have implications for DoD applications is the presence of a force.com platform “Recycle

Bin”, which stores deleted data for up to 30 days, during which the data is available for

restoration. It is unclear whether the platform implements secure deletion for data stored

in the force.com datacenter, and whether there is a mechanism for ensuring that deleted

data is removed from backups.

9. Vendor Assessment: Pervasive Software (PaaS/SaaS)

9.1. Description

Similar to Boomi but wider and more differentiated, Pervasive offers a an on-demand

integration suite for Software-as-a-Service and Platform-as-a-Service (confusingly

renamed Integration-as-a-Service, or IaaS). Pervasive emphasizes development speed,

application integration, as well as heterogeneous data connectivity: (200+ connectors,

among them to legacy COBOL , QSAM and MVS formats).

In addition, its product line field a remarkable capability to process non-relational, semi-

structured and un-structured content, which is typically not explicitly formulated and

buried in most organizations. Specialized pre-made turn key integration solutions for

industry sectors are offered, as well. As with Boomi, hosting options are available (data

integration through Pervasive's DataCloud) or any other cloud, as well as local premises.

Thus, a full range of SaaS-to-SaaS, on-premises-to-SaaS and on-premises to on-premises,

as well as traditional B2B and large-scale bulk data exchange.can be integrated via

Pervasive's platform

Page 26: Security Assessment of Cloud Computing Vendor Offerings

25

Figure 4 Pervasive Data Integration Platform

Their flagship product, the Pervasive Data Integration Platform, consists of a unified set

of tools for rapid development of seamless connectors that captures, as in Boomi's case,

process logic, data mapping and transformation rules. The connector 'secret sauce' lies in

the runtime Integration Engine, instantiated by the Integration Services (as shown in blue

and purple in Figure 4)

9.2. Security Assessment

In contrast to Boomi's strong emphasis on security (attested to by several position and

technical papers), surprisingly little detail is found on Pervasive's security stance. No

security controls are mentioned for the DataCloud. One sentence in their technical

description of the Integration Engine is devoted to noting that each Integration Engine

instantiated runs in its own address space, their isolation increasing reliability .

Several issues can be deduced from the type and scale technologies employed: The

Management of deployed component though the Integration Manager (yellow in Figure

4) is effected through standard browsers - this is subject to the standard XSS and CSFR

issues as delineated in previous section. Lastly, the workhorse of their integration

platform, the Integration Engine is designed to handle an impressive gamut of

applications (as evinced from Figure 5), from extract, transform, and load (ETL) projects

to B2B integration (e.g. HIPAA, HL7, ACORD, X12 and SOA adapters).

This in itself is not the problem, but coupled with the Engine being lightweight to deploy,

as Pervasive emphasized several times, thorough input validation seems unlikely. Before

any adoption for mission critical deployment, it is recommended that these and other

Page 27: Security Assessment of Cloud Computing Vendor Offerings

26

issues (e.g. interaction of applications that embed the Integration Engine is accomplished

also through relative complex COM APIs) be addressed.

Figure 5: Integration Engine API

Despite several contacts via email, telephone conversation with sales people, account

managers and senior system engineers with entreaties for material addressing the security

issues as formulated by the Navy's SOW, no further material was forthcoming as of date

of this writing.

We stress that the paucity of available information needn't necessarily reflect on the

quality of their safety controls: Indirect evidence that security controls across their IaaS,

PaaS and SaaS must meet minimum standards can be corroborated by 100+ case studies,

which span over a dozen sectors. Pervasive's family of products have been deployed as an

integration solution in industries subject to comparable information classification, audit

and access control stipulations as the Navy; health care sector being one of them.

9.3. Case Studies of Interest

The health care sector is of interest because of its statutory (HIPAA) data security

requirements, process complexity, entity scale and legacy system integration. Pervasive

lists about a dozen and a half case studies, among them the State of New York. NY

decided to modernize its Medicaid system with the Pervasive Data Integrator for

efficiency and HIPAA compliance reasons.

Originally created to streamline healthcare processes and reduce costs, HIPAA mandates

minimum standards to protect and keep private an individual‟s health information. For

organizations to be compliant, they must design their systems and applications to meet

Page 28: Security Assessment of Cloud Computing Vendor Offerings

27

HIPAA‟s privacy and security standards and related administrative, technical, and

physical safeguards.

These standards are referred to as the so-called Privacy and Security Rules. HIPAA‟s

Privacy Rule requires that individuals‟ health information be properly protected by

covered entities. Among other requirements, the privacy rule regulates encryption

standards for data in transmission (in-flight) and in storage (at-rest). HIPAA's Security

Rule mandates detailed administrative, physical and technical safeguards to protect health

information: Inter alia, this means implementation and deployment of access controls,

encryption, backup and audit controls for the data in question, subject to appropriate

classification and risk levels. Other industry case studies of potential interest to the Navy

may involve Transportation/Manufacturing (logistics), Public Sector/ Government

(statutes) and Financial (speed) sectors.

Judicious study of health care case studies may also yield insights into issues of scale and

legacy system migration. In this context, we mention the state of California's tackling of

HIPAA compliance with Pervasive software. Their unique requirements - non-negotiable

integration of legacy systems and a traffic volume of 10 million transactions/month have

Navy correspondences. Lastly, Louisiana East Jefferson General Hospital's transition

from proprietary database ETL tool to a Pervasive solution in order to optimize use of

their data warehouse may warrant a look, as well.

Page 29: Security Assessment of Cloud Computing Vendor Offerings

28

Author Short Bio: Vassil Roussev

Vassil Roussev is an Associate Professor of Computer Science at the University of New

Orleans (UNO). He received his Ph.D. in Computer Science from the University of North

Carolina—Chapel Hill in 2003. After that, he joined the faculty at UNO and has focused

his research into several related areas—computer security, digital forensics, distributed

systems and cloud computing, high-performance computing, and human-computer

interaction. The overall theme of his research is to bring to bear the massive

computational power of scalable distributed systems, as well as visual analytics tools to

solve challenges in security and forensics with short turnaround times. He is also working

on tighter integration of security and forensics tools as a means to enrich both areas of

research and practice.

Dr. Roussev‟s has over 20 peer-reviewed publications (book chapters, journal articles,

and conference papers) in the area of computer security and forensics, including featured

articles in IEEE Security and Privacy and Communications of the ACM. His research and

educational projects have been funded by DARPA, ONR, DoD, SPAWAR New Orleans

the State of Louisiana, and private companies, including a Sun Microsystems Academic

Excellence Grant.

Dr. Roussev is Director of the Network Security Lab (NSSAL) at UNO, coaches the

UNO Collegiate Cyber Defense Team, and represents UNO at the Large Resource

Allocations Committee of the Louisiana Optical Network Initiative (http://loni.org). He is

also a Co-PI on a $15mln project to create the LONI Institute (http://institute.loni.org/).

The LONI Institute seeks to develop a state-wide collaborative R & D environment

among Louisiana‟s research institutions with a clear focus on advancing computational

scientific research.

Page 30: Security Assessment of Cloud Computing Vendor Offerings

29

Author Short Bio: Golden G. Richard, III

Golden G. Richard III is Professor of Computer Science in the Department of Computer

Science at the University of New Orleans. He received a B.S. in Computer Science (with

honors) from the University of New Orleans in 1988, and an M.S. and Ph.D. in Computer

Science from The Ohio State University in 1991 and 1995, respectively. He joined UNO

in 1994. Dr. Richard‟s research interests include computer security, operating systems

internals, digital forensics, and reverse engineering. He is a GIAC-certified digital

forensics investigator and a member of the ACM, IEEE Computer Society, USENIX, the

American Academy of Forensics Sciences (AAFS), and the United States Secret Service

Taskforce on Electronic Crime. At the University of New Orleans, he directs the Greater

New Orleans Center for Information Assurance and co-directs the Networking, Security,

and Systems Administration Laboratory (NSSAL).

Prof. Richard has over 30 years of experience in computing and is a recognized expert in

digital forensics. He and his collaborators and students at the University of New Orleans

have made important research contributions in high-performance digital forensics, file

carving, evidence correlation mechanisms, on-the-spot digital forensics, and OS support

for digital forensics. Furthermore, he and his collaborators pioneered the use of Graphics

Processing Units (GPUs) to speed processing of digital evidence. Recently, he developed

and taught one of the first courses in academia on reverse engineering of malicious

software. He is the author of numerous publications in security and networking as well

as two books for McGraw-Hill, the first on service discovery protocols (Service and

Device Discovery: Protocols and Programming, 2002) and the second on mobile

computing (Fundamentals of Mobile and Pervasive Computing, 2005).

Page 31: Security Assessment of Cloud Computing Vendor Offerings

30

Author Short Bio: Daniel Bilar

Education

Dartmouth College (Thayer School of Engineering), PhD Engineering Sciences, 2003 Thesis: Quantitative Risk Analysis of Computer Networks

Cornell University (School of Engineering), MEng Operations Research and Information Engineering, 1997

Brown University (Department of Computer Science), BA Computer Science, 1995 Current Affiliation

Assistant Professor of Computer Science, University of New Orleans, August 2008 - present

Co-Chair, 6th Workshop on Digital Forensics and Incident Analysis (Port Elisabeth, SA), 2010

Advisory Board, Journal in Computer Virology (Springer, Paris), 2008-

Professional Advisory Board, SANS GIAC Systems and Network Auditor, 2002-2005 Past Affiliations

Endowed Faculty Fellow, Wellesley College (Wellesley MA), 2006-2008

Visiting Professor of Computer Science, Colby College (Waterville ME), 2004-2006

Research Interests Detection, Classification and Containment of Highly Evolved Malicious Software, Systems of Systems Critical Infrastructure Modeling and Protection, Risk Analysis and Management of Computer Networks Dr Bilar was a founding member of the Institute for Security and Technology Studies at Dartmouth College, conducting counter-terrorism technology research for the US Department of Justice and Department of Homeland Security.

Page 32: Security Assessment of Cloud Computing Vendor Offerings

31

List of Abbreviations

CC Common Criteria for Information Technology Security Evaluation

DBMS Database management system

DoD Department of Defense

EAL Evaluation assurance level

EC2 Elastic cloud computing

IaaS Infrastructure as a service

IT Information technology

HIPAA The Health Insurance Portability and Accountability Act of 1996

LAMP Linux, Apache, MySQL, Perl/PHP/Python

NIAP National Information Assurance Partnership

OS Operating system

PaaS Platform as a service

PR Public relations

SaaS Software as a service

SLA Service level agreement

TCO Total cost of ownership

TPM Trusted platform module

Page 33: Security Assessment of Cloud Computing Vendor Offerings

32

References

[1] Amazon.com, “Creating HIPAA-Compliant Medical Data Applications with

AWS”, April 2009,

http://awsmedia.s3.amazonaws.com/AWS_HIPAA_Whitepaper_Final.pdf.

[2] Amazon.com, “Amazon Web Services. Customer Agreement”,

http://aws.amazon.com/agreement/.

[3] Amazon.com, “Amazon Web Services: Overview of Security Processes”, Sep 2008,

http://s3.amazonaws.com/aws_blog/AWS_Security_Whitepaper_2008_09.pdf

[4] Nicholas Arvanitis, Marco Slaviero, Haroon Meer, "Clobbering the Cloud", BH

USA 2009, August 2009, http://www.sensepost.com/research/presentations/2009-

08-SensePost-BH-USA-2009.pptx

[5] Adam Barth, Collin Jackson and John C. Mitchell, "Robust Defenses for Cross-Site

Request Forgery", Proceedings of the ACM CCS 2008, October 2008,

http://www.adambarth.com/papers/2008/barth-jackson-mitchell-b.pdf

[6] Boomi Inc, "Boomi OWASP Top Ten Response", August 2009,

http://www.boomi.com/files/boomi_datasheet_owasp_response.pdf

[7] Rajkumar Chee, Shin Yeo, and Srikumar Venugopal, "Market-Oriented Cloud

Computing: Vision, Hype, and Reality for Delivering Computing as the 5th

Utility", Proceedings of the 2009 9th IEEE/ACM International Symposium on

Cluster Computing and the Grid, May 2009

[8] DARPA Microsystems Technology Office, BAA 07-24 "TRUST in Integrated

Circuits", March 2007, http://www.darpa.mil/MTO/solicitations/baa07-

24/index.html

[9] DOD Defense Science Board Task Force, "High Performance Microchip Supply",

Feb 2005, http://www.acq.osd.mil/dsb/reports/2005-02-HPMS_Report_Final.pdf

[10] DoD DISA, "Security Checklists", http://iase.disa.mil/stigs/checklist/index.html

[11] eSecurity Planet, “Salesforce.com Scrambles To Halt Phishing Attacks”,

http://www.esecurityplanet.com/trends/article.php/3709871/Salesforcecom-

Scrambles-To-Halt-Phishing-Attacks.htm.

[12] Gartner, Inc. “SaaS CRM Reduces Costs and Use of Consultants” by Michael

Maoz. 15 October 2008

[13] Panagiotis G. Ipeirotis, Luis Gravano, Mehran Sahami, “Probe, Count, and

Classify: Categorizing Hidden Web Databases”, ACM SIGMOD 2001, Santa

Barbara, California, USA

[14] McKinsey & Co., “Report: Clearing the Air on Cloud Computing”, Apr 2009,

http://uptimeinstitute.org/content/view/353/319/.

[15] NDP LLC, "Why is SAS 70 Relevant to SaaS in Today's Regulatory Compliance

Landscape?", 2009, http://www.sas70.us.com/industries/saas-and-sas70.php

[16] Thomas Ristenpart, Eran Tromer, Hovav Shacham and Stefan Savage, "Hey, You,

Get Off My Cloud! Exploring Information Leakage in Third- Party Computer

Clouds", Proceedings of ACM CCS 2009, Nov. 2009,

http://cseweb.ucsd.edu/~hovav/dist/cloudsec.pdf

Page 34: Security Assessment of Cloud Computing Vendor Offerings

33

[17] salesforce.com, “ISO 27001 Certified Security”,

http://www.salesforce.com/platform/cloud-infrastructure/security.jsp.

[18] salesforce.com, “Three global centers and disaster recovery”,

http://www.salesforce.com/platform/cloud-infrastructure/recovery.jsp.

[19] Nuno Santos, Krishna P. Gummadi, and Rodrigo Rodrigues, “Towards Trusted

Cloud Computing”, USENIX Workshop on Hot Topics in Cloud Computing, San

Diego, CA, Jun 2009.

http://www.usenix.org/events/hotcloud09/tech/full_papers/santos.pdf

[20] Alex Stamos, Andrew Becherer and Nathan Wilcox, "Cloud Computing Models

and Vulnerabilities - Raining on the Trendy New Parade", Blackhat USA Briefings,

July 2009, https://media.blackhat.com/bh-usa-09/video/STAMOS/BHUSA09-

Stamos-CloudCompSec-VIDEO.mov

[21] UC Berkeley Reliable Adaptive Distributed Systems Laboratory, "Above the

Clouds: A BerkeleyView of Cloud Computing", Feb 2009,

http://radlab.cs.berkeley.edu/

[22] Luis M. Vaquero, Luis Rodero-Merino, Juan Caceres, Maik Lindner, “A Break in

the Clouds: Towards a Cloud Definition”, ACM SIGCOMM Computer

Communication Review, Volume 39, Number 1, January 2009.

[23] United States Government Accountability Office. “INFORMATION

ASSURANCE National Partnership Offers Benefits, but Faces Considerable

Challenges”, Mar 2006, http://www.gao.gov/new.items/d06392.pdf.