Download - Toigo Critical Convergence
Critical Convergence
Presented byJon William Toigo
CEO Toigo Partners InternationalFounder The Data Management Institute
Copy
right
© 2
009
by T
oigo
Par
tner
s In
tern
atio
nal L
LC.
All
Righ
ts R
eser
ved.
Convergence /kən-vər-jən(t)s/
• Noun. L.L. convergere "to incline together" from com- "together" + vergere "to bend"
1. The act of converging and especially moving toward union or uniformity ; especially coordinated movement of the two eyes so that the image of a single point is formed on corresponding retinal areas
2. The state or property of being convergent
3. Independent development of similar characters (as of bodily structure of unrelated organisms or cultural traits) often associated with similarity of habits or environment
4. The merging of distinct technologies, industries, or devices into a unified whole
5. Mathematics The property or manner of approaching a limit, such as a point, line, function, or value.
Great Marketing Term
• Can be loaded with any value you want
• Three meanings I want to explore– Convergence of front and back office goals
and strategies
– Convergence of IT architectural models
– Convergence of drivers around intelligent data management
• These constitute “critical convergence” areas…
“Critical” Convergence?
• “Critical” connotes normative value: the way things ought or need to be…– Business and IT need to be realigned in most
companies for productive work to be done
– IT infrastructure needs to be purpose-built to business requirements and components and processes must be commonly manageable
– Data assets need to be handled much more intelligently and granularly than they are today
• Says who? Trust me, it isn’t only me.
Critical Convergence Point 1:Mending the Front/Back Office Rift
Situation Today in Too Many Firms• Current thinking: IT costs too much with no
demonstrable return on investment– Committees formed to prioritize purchases
(often without IT representation)
– TCO hammer makes all IT costs look like nails
• Leading to frantic adoption of “silver bullet” methodologies: ITIL, ISO, CoBIT, SOA, etc.
• Generating more frustration at the top…and bottom of the organization
– CIO office has a revolving door
– Laissez faire attitudes regarding product suitability – “It’s not my money.”
– Uptick in outsourcing/in-sourcing – despite fact that a problem cannot be outsourced
Speaking Different Languages
• The Front Office focused on– Cost-Containment
– Compliance & Governance
– Continuity & Risk Management
– Top Line Growth
• The Back Office focused on– Infrastructure Efficiency
– Performance Management
– Service Level Improvement
– New Information Products
Current “C-4” Challenges Can (and Should) Unite Us
• The four issues that are front of mind for business IT planners:
Not Just a Reflection of the Current Economy…
• The C-4 issues have been building steadily for decades, a function of…1. Failure to “purpose-build”
infrastructure…especially in the distributed computing environment
2. Failure to instrument infrastructure for monitoring, measurement and management
3. Failure to manage data effectively
The current recession has only underscored the problems and, perhaps, given us a “pause” to think strategically about ways for solving them…
The Front and Back Office Need to Converge
• Common language needed to describe – Challenges and opportunities
– Strategic vision and goals
– Objectives and success metrics
• More effective division of labor– Business value case development
– Product selection
– Implementation plans
• In an atmosphere of mutual respect…
Critical Convergence Point 2:Returning to Purpose-Built Infrastructure
Back to Basics• What is “purpose-building”
infrastructure?– n. a kind of “intelligent design” – a deliberate
and well conceived matching of infrastructure “services” (resources and functions) to meet well-defined business needs…
– v. the act of buying or developing and deploying technology that serves a business need and not simply outsourcing your thinking process to vendors whose top line growth is more important than your top line growth
NOT a New Idea:Data “DNA” Inherited from Business Processes and
Applications
“Data Hosting” Specification Should Dictate Resources and Services
Geez, that sounds like
DP-101class…
Chances are Good That Your Infrastructure Was Originally Built That Way…
• Might have started with a data-centric, business process-focused view, but it rarely stayed that way over the years– Management changed: Different masters bring
different views…
– Technology changed: Distributed systems, TCP/IP, the Internet, the “V” word, et al…
– Business changed: Mergers and acquisitions, new products, new go to market strategies…
– Attitudes changed: “IT is our strategic differentiator,” “Bricks and Sticks are Dead,” “Does IT Matter?,” “Legal is running IT now,” “Web 2.0,” “Enterprise 2.0,” “Mash-ups will replace IT applications,” Clouds will replace data centers, blah blah blah…
Biggest Problems Attributable to Triumph of Marketecture
• Marketecture: n. that which otherwise good architecture becomes when subjected to vendor marketing departments; see kluge– The rise of distributed systems: mainframe
deconstruction gone awry…
– The network is the computer: distributed computing gone awry…
– SANs: storage infrastructure gone awry
– x86 server virtualization: LPARs gone awry
– “Cloud computing:” multi-tenancy gone awry
Purpose-Built Tech
“Cool” Tech
MA
RK
ETI
NG
Marketecture Myths Abound
• “IT isn’t about building and managing infrastructure anymore, it’s about managing a few trusted vendors…”
• “Smarter systems reduce management burden (and its associated labor costs) and fit the needs of all businesses well…”
• “Resource utilization efficiency is best achieved through server virtualization…”
• “Why deal with problems? Source your services from a cloud…”
And They Resonate Because…• Leading vendors tend to sell to the
front office, not to the back office (where IT lives)…– Eschewing “tech speak” and framing pitch
in terms of business problem solving
– Purchasing mind share with PPV analyst reports
– Casting aspersions on internal IT competency
– Plus, the usual schmoozing…
Not to Suggest that IT is Blameless…
Latest Challenges to Purpose-Built
• Return to the “mainframe-centric” model of the universe
• Rise of the smart storage array
• Hypervisors as generic operating systems
• Standards-free “Cloud Computing” paradigms
Mainframes and Mini-Me’s• Mainframes enjoying long-deserved
renaissance– Workload shifting to mainframes like never
before
– Most companies deferring or reversing MF decommission plans
• Prompting the appearance of mainframe mini-me’s – Dedicated proprietary hardware stovepipes
– De facto standards
– Proprietary software protocols for interconnecting bus
Fundamental Question:What is the Proper Order of the IT Universe?
• Mainframe centric
• Network centric
• Storage centric
Data centric
So, Is “Purpose-Built” a Matter of Interpretation?
HP claims that its Oracle platform is purpose-built: to host an Oracle database (Q: Are all Oracle DBs the same?)
EMC claims that NetApp makes a mistake by building one-size-fits-all storage platforms, which they have learned do not meet the needs of every application: hence, they offer at least four storage product lines, calling this a purpose-built strategy (Note: most value-add functions require homogenous EMC infrastructure)
Xiotech has a Lego™ building block storage array that works with any controller and lacks embedded software functionality found in “value-add” arrays: Cobble together various ISE bricks and use software components
from third party vendors in their ecosystem to cobble together exactly the capacity + software functionality set you need
Commonly manageably using open W3C Web Services standards…
Storage is Key to Purpose-Built Design
• 33 to 70 cents of every dollar spent on IT hardware today
• 300% capacity growth expected between 2007 and 2011
• Biggest power pig in the data center
• Where data goes to sleep…
Data Re-Reference Rates Already Baked Into Some Product Categories
Prob
abili
ty o
f Reu
se100
0
“Primary Storage”
“Secondary Storage”Accesses to Data
0 days 30+ days 90+ 1 yr ∞Time
milliseconds seconds minutes hours daysAccessSpeed
“Retention Storage”“Tertiary Storage”
Embedded Value-Add Software: A Good Idea?
• The array controller is getting bloated– Certain functions need to be performed close
to the spindles themselves; most don’t
– Dynamic established in market: must add new “value” to controllers every six months or you look like you are standing still
– Vendors argue pre-integration value; stifle open discussion of performance deficits
• Driving up cost– You pay for all functionality, whether you use it
or not
– Management is obfuscated: need to hire more monks when you buy a new array
One Practical Consequence Worth Noting
1980 2010Sub-MB
TB+
CAPA
CITY
(Are
al D
ensi
ty =
Gb
per
in2 )
COST PER G
B
HIGH
LOW
ARRAY COSTS ACCELERATING
Source: “Avoiding the Storage Crunch,”Jon Toigo, Scientific American
Placed in a SAN, Now You’re Talking Real Money
• SANs aren’t networks: ENSA has never been brought to market
• SAN switch ports remain pricey, underutilized (sub 15% in most cases), accelerating price for primary storage to as high as $150/GB, secondary storage to $110/GB, tertiary to $50-$80/GB
• At those prices, and anticipated capacity growth…
More Bad News• Server virtualization isn’t the
big fix everyone is hoping for…
• Nothing wrong in concept: mainframe LPARs and PRISM around for a couple of decades
• But x86 extent code for multi-tenancy not fully ripened
• And hypervisors confront a two-front war…
A War on Two Fronts…
• Good hypervisors can do a great job with x86 extent code…
• The challenge to stability…– From above: applications
making “illegal” resource calls
– From below: value-add functionality proffered by storage vendors (thin provisioning, for one) can compromise expected resource request fulfillment
App software vendorsdon’t write code to support VM hosting…
Many storage vendors are“adding value” by obfuscating
actual capacity…
Latest Elephant in the Room
• Storage vendors cozying up to server virtualization vendors using storage resource utilization enhancement speak:
– “3PAR has worked with VMware on its new adaptive queue depth algorithm… dynamically adjusts the LUN queue depth in the VMkernel I/O stack to minimize the impact of I/O congestion.”
– But, “Queue depth is just one element which regulates what traffic is allowed onto the "SAN Freeway" but the entire SYSTEM is what needs to be balanced AND MEASURED… NetWisdom is like the ‘Eye in the Sky’ that sees everything. We collect 65 metrics. Latency is very important.”
• Two effects:– More hype about storage management advantages of x86
virtualization, stalling work on real solutions
– Multiple proprietary attachment and I/O load balancing/routing APIs inspiring abandonment of networked storage vision
Is Server Virtualization Really Ready to be the Cloud OS?
• This week’s trend or a valid model?– Service bureau computing begat mainframe
multi-tenancy begat ASPs begat Cloud storage…
– Electronic vaults begat SSPs…
• Questions abound:– Can cloud providers do storage more
efficiently than you can?
– Can cloud provider SLAs be trusted?
– Is data secure in a cloud?
– Do clouds present governance issues?
– Why is man born only to suffer and die?
Purpose-Built Infrastructure May Embrace Clouds Someday…
Realizing a Purpose-Built Infrastructure Requires…
• “Deconstructionalist” thinking– Willingness to let go of stovepipe vendors
– Willingness to cobble together required services from best of breed suppliers rather than one-stop-shop providers
– Desire to achieve best efficiency at lowest cost
• Embrace of common, standards-based approach to managing of infrastructure
• Understanding of data hosting requirements and intelligent routing of I/O across infrastructure (we will get to this one)
A Quick Historical Perspective
• Tendency to view mainframe storage management as inherently superior to distributed systems SRM
• In the late 1970s, IBM blessed the world with SMS and HSM, bringing order to the mainframe storage universe
– Mainframe storage allocation efficiency is 4x that of distributed storage allocation efficiency
– Mainframe storage utilization efficiency is 3-4x that of distributed storage
Mainframe Storage CAE: ~80%*Distributed Storage CAE: ~20%
Mainframe Storage CUE: ~70%**Distributed Storage CUE: ~17-20%***
*Assumes SMS**Assumes HSM***Depends on Server OS
IBM’s Contribution Was a Blessing
• But it was also limited to IBM architecture…– IBM Cosmology: Mainframe at the center of
the IT universe• Leveraged de facto standards and well-
defined architectural model
• Homogeneity: all storage must conform to IBM rules
• Sets up a rigid workflow model: data goes from system memory to DASD to tape (or virtual tape) over time
– Actual goal: sell more DASD, while maintaining sparse labor force
HSM Added Value: Classes for Policy-Based Data Movement
STORAGEADMINISTRATOR
END USERS APPLICATIONS
STORAGE GROUP
Good Enough Then, but Now?
• Mainframe world in transition– More, varied workloads
– Different data management needs
– Consolidation of servers into LPARs
• However, SMS/HSM have not fully transitioned into the z-Series world– Still part of z/OS, but not part of z/Linux
– HSM does not extend to z/Linux environments
Open Systems Corollaries• “Trust your vendors to solve your management
problems.”
• Smarter arrays do it all– Capacity management through embedded tiering
algorithms
– Capacity management through embedded thin provisioning algorithms
– Capacity management through embedded compression and block de-duplication algorithms
• Hypervisors on the server allocate storage resources to apps-in-VMs elegantly– The rise of mainframe wannabes
– The rise of cloud storage
“See that green light? I know I’ve got a problem if it turns red.”
• Primed to see things this way from early on…– Common view in small shops with only one storage platform from
one vendor
– Indicator lights, self-articulating web pages with status and configuration data, or a utility CD provides a solid “storage management” story
– Meets the standard requirements of storage management Provides configuration control
Monitors device status and capacity
Reports when problems occur so they can be addressed
Easy to use, ready out of the box, no muss, no fuss
But Element Managers Become a Problem When…
• Volume of data grows…
• The number of storage products increases…
• The storage becomes heterogeneous…– Products deployed from different vendors
– Different products deployed from the same vendor
• Storage becomes “smart”…– Vendors build “spoofs” into gear to improve performance
(example: NVRAM caching)
– Array controller-embedded “value-add” software obfuscates real monitoring of capacity
But On-Array Thin Provisioning will Take Care of All That
• A storage capacity “shell game”– Leverage “reserved-but-not-allocated”
space to over-allocate disk capacity
– No standards, only as effective as disk space demand forecasting algorithm
– Can defer need to buy more storage
– But data at risk of a “margin call”
• Not a replacement for storage management…
Thin is In, or Is It?Survey of 249 Companies
• 79% expect TP to deliver improved capacity allocation efficiency
• 49% expect easier provisioning with less disruption
Conversely,• 44% concerned that TP will result in greater
risk of running out of storage• 43% worry that TP adds complexity• 42% worry about a lack of capacity
management tools
Okay. So, Compression and De-Dupe: That’ll Fix It.
• Quick definition: reduction in the number of bits to represent data; can also mean:
– Removing/stubbing bit patterns
– Storing only change bytes or bits in versioning system
– Elimination of file duplicates
• Ask a vendor about the differences and make some coffee…
• Purported Management value: – Squeeze more data onto media
– Defer need to buy more capacity
• Issues: – Your mileage may vary
– Possible compliance issues
– More equipment to power
• Over time, even a trash compactor fills a landfill…
• Replacement for storage management?
The new Marlboro 72s may besmaller than old Reds, and lessexpensive to buy, but they willeventually kill us just the same…
Ahh, But Server Virtualization is Coming to the Rescue
• Leveraging x86 chip code extents to stack virtual machines in a single server chassis…– To consolidate assets
– To improve resource allocation efficiency (abysmally poor in distributed systems)
– To simplify storage management and automate data movements (hypervisor controls storage)
• Hmm. We talked about this already at some length, but…
The Simple Truth
• Good stuff: x86 Hype has focused attention on need for storage management from a systemic viewpoint
• Problems• Industry resists common virtualization
standards: x86 chip extensions alone are not enough
• Even if Hypervisor is “well behaved,” not necessarily the case with application software: performance issues result
• Storage vendors keep throwing curves, destabilizing VM stack from below
Plus, the Problem of “Dark Storage”
• Early on, storage associated with VMs was double-counted…since resolved by leading SRM tool makers
• Now, we have the problem of “dark storage”– A TB allocated to server admin
– Server admin applies file system to 500 GB, forgets about his “reserved capacity”
– Storage admin believes 1TB allocated, Server admin believes he has 500 GB – 500 GB dark and beyond detection
• Virtualization doesn’t simplify storage management, it adds an abstraction layer that can confuse and obfuscate capacity management and data management…
What About All of those SAN Management Standards?
• Hardware and cabling– FC Standards (T-11 ANSI) defined by and provide
wiggle room for vendors
– No guarantee of interoperability: switch/HBA/controller firmware upgrades can make the SAN disappear
• SMI-S from SNIA– A bust: implemented on a fraction of hardware
platforms
– Better management data from APIs and SNMP MIBs even when hardware is SMI-enabled
• Kind of feels like the storage industry’s heart isn’t in it…
Lack of Strategic Management in Distributed Systems is Lagging
• Wild West World of Storage Standards combined with Vendor Proprietary Value-Add designs in a Heterogeneous Environment equals…– Lack of visibility and huge resource waste
– No means to anticipate problems or requirements
– No effective means to police vendor claims
– Increase in storage-related downtime
– More time spent fire-fighting tactical issues
• Issues creep up over time: Frog Boiling in a Pot
Contributing Significantly to TCO
Backup and Recovery
Hardware Management
ScheduledDowntime
Administration
Environmental
Hardware Acquisition
30%
3%
20%
13%
~14%
20%
COST TO
MANAGE
COST TO
ACQUIRE
Administration (labor) and other
“soft costs” equal to 4x
acquisition costsannually.
Acquisition cost is only 20% of
TCO
Source: Gartner
Bottom Line• Purpose-building infrastructure to realign IT
with business will not be easy– May need to abandon “comfort zones” with
“love brands”
– May need to revisit build vs buy decisions
– Decide on management paradigm first, and promote management via preferred approach to a primary selection criteria
– No forklift upgrades: phase in as part of normal hardware refresh
– Work with an integrator who isn’t just an order taker for brand name vendors
• It starts with a strategic vision and some business-savvy…
My Preferred Paradigm: Business-Centric, Standards-Based and Data-Focused
• Data focused? Effective infrastructure management is required to get to the Holy Grail -- Data Lifecycle Management -- across purpose-built storage infrastructure
– For real D/ILM to happen, storage classes need to be defined and operational status monitored continuously
– Selecting or building storage using heterogeneous components and fielding services on networked platforms (aka appliances, routers, gateways, etc.) makes management a must-have
• W3C Web Services-based management support by gear vendors would make for an easily deployed and integrated enterprise management solution (covering servers, networks and storage)
SOAP/XML could provide the “universal glue” required to auto-
provision and auto-protect applications with infrastructure-based
resources and services.
Already supported by a broad range of applications, OS software, network
hardware…and now storage.
Critical Convergence Point 3:Intelligent Data Management
Intelligent Data Management is No Longer Optional
• Data Management: “The undiscovered country”
• Absolutely essential to – Align business processes with IT investments and to
demonstrate business value
– Achieve sustainability in service-oriented strategy
– Realize capacity utilization efficiency and contain costs
– Comply with regulatory mandates
– Protect data effectively
– Green IT operations
In the Absence of Data Management
• Storage is a junk drawer – as much as 70% of capacity is wasted
• After normalizing over 10,000 storage assessments, we find– 30% of the data occupying spindles is
useful
– 40% is inert (needed for retention, but never accessed)
– 15% of capacity is allocated but unused (often “dark storage”)
– 10% is orphan data whose owner/server no longer exist
– 5% (low estimate) is contrabandSource: Randy Chalfant, CTO, Sun Microsystems, from Making IT Matter (Chalfant/Toigo, 2008)
Sorting the Storage Junk Drawer
• Conceptually, a straightforward undertaking
1. Classify your data
2. Create policies for data handling by class
3. Instrument your infrastructure for policy-based data movement
4. Move your data around per policy
• Then, as in writing, just open a vein and bleed on the page…
Classification requires Information Collection and Analysis
• Technically speaking, you need to perform a business process deconstruction analysis
– Start with a few key business processes: senior management will probably be able to point you to the right ones
– Deconstruct processes into tasks and tasks into workflows, then identify data associated with a workflow
– Identify data handling requirements associated with each workflow’s data (retention requirements, criticality, disposal requirements, encryption requirements, etc.)
– Record everything (a spreadsheet works well)
• (This is a wildly profitable service offering for solution providers…)
BUSINESS PROCESS 1
TASK 1.1
TASK 1.2
TASK 1.n
WORKFLOW 1.1.1
WORKFLOW 1.1.2
WORKFLOW 1.1.n
DATA
DATA
DATA
DATA
Classification Simplified
Data NameWorkflow ID
Updat
e Fr
eque
ncy
Acces
s Fre
quenc
ySt
ale
bySp
ecia
l Ret
entio
nDel
etio
n Re
quire
men
t
Refe
renc
e Dat
a?Cr
itica
lity
Secu
rity
Other
Req
uire
men
tOth
er R
equi
rem
ent
Data NameWorkflow ID
Updat
e Fr
eque
ncy
Acces
s Fre
quenc
ySt
ale
bySp
ecia
l Ret
entio
nDel
etio
n Re
quire
men
t
Refe
renc
e Dat
a?Cr
itica
lity
Secu
rity
Other
Req
uire
men
tOth
er R
equi
rem
ent
Just sort the spreadsheetand basic data classes
separate out like a parfait…
Shampoo, Rinse, Repeat• Often heard complaints…
– “What if there are a lot of tasks, workflows and data inputs and outputs?”
– “I have thousands of business processes: you really think we can do this in a timely way?”
– “Jon, this is your classic waterfall approach, which has already come under criticism in fields like application development. Isn’t it easier just to backup/mirror/encrypt everything – all on disk drives?“
– Why is man born only to suffer and die?
• OMG, you folks complain a lot…
But I Feel Your Pain…• Truth be told, there aren’t a lot of short cuts
• Hazards of NOT classifying data by business process-related criteria include– Application of inappropriate or overly expensive
data hosting and protection strategies to relatively unimportant or low priority data assets
– Conversely, the failure to include mission critical data assets in protection schemes
– Lack of predictable time-to-data outcomes from ANY data protection scheme
– None of the extra business value that should derive from data analysis
• Any way you cut it, the results can’t be good
A Few Tools that Can Help• Tek-Tools Storage Profiler
– Lightweight deployment
– Good storage assessment reporting for identifying dupes and dreck
• Digital Reef – First “deep blue math” e-discovery algorithm that
works
– Aimed at information governance and storage capacity reclamation
• NSM (Novell Storage Manager)– Classify data by user profile
– Microsoft Active Directory support improving
– Robust reporting on the way
Just a Gol’ Dern Second…• You are talking about an “SRM” tool (Tek-
Tools), an information governance tool (Digital Reef) and a file lifecycle management tool (Novell): what has that got to do with a C-4 Strategy?
• Each tool provides the means to – Locate data assets on infrastructure
– Cull out the junk data and facilitate the introduction of intelligent data management to develop policy-based schemes for classifying data for data protection service provisioning
– Simplify manual classification activity, especially with respect to user files, which are the largest component of overall data stored by most companies today
Data Modeling in Action
Outage Costs
Detailed Summary
Regulatory Reqs:(e.g., Privacy, Preservation,
Protection,Discovery)
Criticality
Interdependencies
Interdependencies
Volume, Volatility, Access Frequency
Hard and Soft Costs
Hard and Soft Costs
Hard and Soft Costs
Hard and Soft Costs
Over Time and With Effort…
My Vision of Data Management Convergence…
Data
Bus
Value-Add Software Functions
Hardware Resources
StatusMonitoring
I/O Routes
Provisioning Policy Engine
Providing the Means to Realize C-4 Nirvana: Automated Data Service Provisioning
ServicesRequested for Data Handling
ServicesStatus
Inventoried
“Route” Established for
Services
Requested Services Applied
to Data
Provisioning and Protection
Policies Applied
App Seeks to Write Data
Already Underway• Toigo Partners International and the Data
Management Institute will shortly launch a C-4 Community Portal to aid in W3C Web Services storage reference architecture design, development and education: You Are Invited!
• Working groups will span technical and nontechnical subjects and will combine consumers, vendors and reseller/integrator perspectives…
• Leading to the establishment of a C-4 Microfactory – a “farmer’s market” web site where Web Services-enabled hardware and software products can be discovered and sourced for integrated solutions…