4/17/2013

1

1

Introduction to Grid Computing

Based on:Grid Intro and Fundamentals Review

Talk by Gabrielle Allen

Talk by Laura Bright / Bill Howe

2

3

Overview

• Background: What is the Grid?• Related technologies• Grid applications• Communities

• Grid Tools• Case Studies

4/17/2013

2

4

PUMA: Analysis of Metabolism

PUMA Knowledge Base

Information about proteins analyzed against ~2 million gene sequences

Analysis on Grid

Involves millions of BLAST, BLOCKS, and

other processesNatalia Maltsev et al.http://compbio.mcs.anl.gov/puma2 4

5

Initial driver: High Energy Physics

Tier2 Centre ~1 TIPS

Online System

Offline Processor Farm

~20 TIPS

CERN Computer Centre

FermiLab ~4 TIPSFrance Regional Centre

Italy Regional Centre

Germany Regional Centre

InstituteInstituteInstituteInstitute ~0.25TIPS

Physicist workstations

~100 MBytes/sec

~100 MBytes/sec

~622 Mbits/sec

~1 MBytes/sec

There is a “bunch crossing” every 25 nsecs.

There are 100 “triggers” per second

Each triggered event is ~1 MByte in size

Physicists work on analysis “channels”.

Each institute will have ~10 physicists working on one or more channels; data for these channels should be cached by the institute server

Physics data cache

~PBytes/sec

~622 Mbits/sec or Air Freight (deprecated)

Tier2 Centre ~1 TIPS

Tier2 Centre ~1 TIPS

Tier2 Centre ~1 TIPS

Caltech ~1 TIPS

~622 Mbits/sec

Tier 0

Tier 1

Tier 2

Tier 4

1 TIPS is approximately 25,000

SpecInt95 equivalents

Image courtesy Harvey Newman, Caltech 5

6

Computing clusters have commoditized supercomputing

Cluster Management“frontend”

Tape Backup robots

I/O Servers typically RAID fileserver

Disk Arrays Lots of WorkerNodes

A few Headnodes, gatekeepers and

other service nodes

6

4/17/2013

3

7

What is a Grid?• Many definitions exist in the literature• Early defs: Foster and Kesselman, 1998

“A computational grid is a hardware and software infrastructure that provides dependable, consistent, pervasive, and inexpensive access to high-end computational facilities”

• Kleinrock 1969:“We will probably see the spread of ‘computer utilities’, which, like present electric and telephone utilities, will service individual homes and offices across the country.”

8

3-point checklist (Foster 2002)

1. Coordinates resources not subject to centralized control

2. Uses standard, open, general purpose protocols and interfaces

3. Deliver nontrivial qualities of service• e.g., response time, throughput, availability,

security

9

Grid Architecture

Autonomous, globally distributed computers/clusters

4/17/2013

4

10

Why do we need Grids?

• Many large-scale problems cannot be solved by a single computer

• Globally distributed data and resources

11

Background: Related technologies

• Cluster computing• Peer-to-peer computing• Internet computing

12

Cluster computing

• Idea: put some PCs together and get them to communicate

• Cheaper to build than a mainframe supercomputer

• Different sizes of clusters• Scalable – can grow a cluster by adding

more PCs

4/17/2013

5

13

Cluster Architecture

14

Peer-to-Peer computing

• Connect to other computers• Can access files from any computer on the

network• Allows data sharing without going through

central server• Decentralized approach also useful for

Grid

15

Peer to Peer architecture

4/17/2013

6

16

Internet computing

• Idea: many idle PCs on the Internet• Can perform other computations while not

being used• “Cycle scavenging” – rely on getting free

time on other people’s computers• Example: SETI@home• What are advantages/disadvantages of

cycle scavenging?

17

Some Grid Applications

• Distributed supercomputing• High-throughput computing• On-demand computing• Data-intensive computing

• Collaborative computing

18

Distributed Supercomputing

• Idea: aggregate computational resources to tackle problems that cannot be solved by a single system

• Examples: climate modeling, computational chemistry

• Challenges include:– Scheduling scarce and expensive resources– Scalability of protocols and algorithms– Maintaining high levels of performance across

heterogeneous systems

4/17/2013

7

19

High-throughput computing

• Schedule large numbers of independent tasks

• Goal: exploit unused CPU cycles (e.g., from idle workstations)

• Unlike distributed computing, tasks loosely coupled

• Examples: parameter studies, cryptographic problems

20

On-demand computing

• Use Grid capabilities to meet short-term requirements for resources that cannot conveniently be located locally

• Unlike distributed computing, driven by cost-performance concerns rather than absolute performance

• Dispatch expensive or specialized computations to remote servers

21

Data-intensive computing

• Synthesize data in geographically distributed repositories

• Synthesis may be computationally and communication intensive

• Examples:– High energy physics generate terabytes of

distributed data, need complex queries to detect “interesting” events

– Distributed analysis of Sloan Digital Sky Survey data

4/17/2013

8

22

Collaborative computing

• Enable shared use of data archives and simulations

• Examples:– Collaborative exploration of large geophysical

data sets

• Challenges:– Real-time demands of interactive applications

– Rich variety of interactions

23

Grid Communities• Who will use Grids?• Broad view

– Benefits of sharing outweigh costs

– Universal, like a power Grid

• Narrow view– Cost of sharing across institutional boundaries

is too high

– Resources only shared when incentive to do so

– Grid will be specialized to support specific communities with specific goals

24

Grid Users

• Many levels of users– Grid developers

– Tool developers

– Application developers

– End users

– System administrators

4/17/2013

9

25

Some Grid challenges

• Data movement• Data replication• Resource management• Job submission

26

Some Grid-Related Projects

• Globus• Condor• Nimrod-G

27

What is a grid made of ? Middleware.

Grid Protocols

Grid Resources dedicatedby UC, IU, Boston

GridStorage

GridMiddleware

Com

putingC

luster

Grid resource time purchasedfrom commercial provider

GridMiddleware

Com

putingC

luster

Grid resources sharedby OSG, LCG, NorduGRID

GridMiddleware

Com

putingC

luster

Grid Client

ApplicationUser

Interface

GridMiddleware

Resource,WorkflowAnd DataCatalogs

GridStorage

GridStorage

• Security to control access and protect communication (GSI)• Directory to locate grid sites and services: (VORS, MDS)• Uniform interface to computing sites (GRAM)• Facility to maintain and schedule queues of work (Condor-G)• Fast and secure data set mover (GridFTP, RFT)• Directory to track where datasets live (RLS)

27

4/17/2013

10

28

Globus Grid Toolkit• Open source toolkit for building Grid systems

and applications

• Enabling technology for the Grid • Share computing power, databases, and other

tools securely online • Facilities for:

– Resource monitoring

– Resource discovery

– Resource management

– Security

– File management

29

Data Management in Globus Toolkit

• Data movement– GridFTP

– Reliable File Transfer (RFT)

• Data replication– Replica Location Service (RLS)

– Data Replication Service (DRS)

30

GridFTP• High performance, secure, reliable data

transfer protocol

• Optimized for wide area networks• Superset of Internet FTP protocol• Features:

– Multiple data channels for parallel transfers– Partial file transfers

– Third party transfers

– Reusable data channels

– Command pipelining

4/17/2013

11

31

More GridFTP features

• Auto tuning of parameters• Striping

– Transfer data in parallel among multiple senders and receivers instead of just one

• Extended block mode– Send data in blocks

– Know block size and offset

– Data can arrive out of order

– Allows multiple streams

32

Striping Architecture

• Use “Striped” servers

33

Limitations of GridFTP

• Not a web service protocol (does not employ SOAP, WSDL, etc.)

• Requires client to maintain open socket connection throughout transfer– Inconvenient for long transfers

• Cannot recover from client failures

4/17/2013

12

34

GridFTP

35

Reliable File Transfer (RFT)

• Web service with “job-scheduler” functionality for data movement

• User provides source and destination URLs• Service writes job description to a database

and moves files• Service methods for querying transfer status

36

RFT

4/17/2013

13

37

Replica Location Service (RLS)

• Registry to keep track of where replicas exist on physical storage system

• Users or services register files in RLS when files created

• Distributed registry– May consist of multiple servers at different sites

– Increase scale

– Fault tolerance

38

Replica Location Service (RLS)• Logical file name – unique identifier for contents of file

• Physical file name – location of copy of file on storage system

• User can provide logical name and ask for replicas

• Or query to find logical name associated with physical file location

39

Data Replication Service (DRS)• Pull-based replication capability• Implemented as a web service• Higher-level data management service built on

top of RFT and RLS• Goal: ensure that a specified set of files exists

on a storage site• First, query RLS to locate desired files• Next, creates transfer request using RFT

• Finally, new replicas are registered with RLS

4/17/2013

14

40

Condor

• Original goal: high-throughput computing• Harvest wasted CPU power from other

machines• Can also be used on a dedicated cluster• Condor-G – Condor interface to Globus

resources

41

Condor• Provides many features of batch systems:

– job queueing

– scheduling policy

– priority scheme

– resource monitoring

– resource management

• Users submit their serial or parallel jobs • Condor places them into a queue• Scheduling and monitoring• Informs the user upon completion

42

Nimrod-G• Tool to manage execution of parametric studies

across distributed computers

• Manages experiment– Distributing files to remote systems

– Performing the remote computation

– Gathering results

• User submits declarative plan file– Parameters, default values, and commands

necessary for performing the work

• Nimrod-G takes advantage of Globus toolkit features

4/17/2013

15

43

Nimrod-G Architecture

Security

44

Grid security is a crucial component

• Resources are typically valuable• Problems being solved might be sensitive• Resources are located in distinct

administrative domains– Each resource has own policies, procedures,

security mechanisms, etc.

• Implementation must be broadly available & applicable– Standard, well-tested, well-understood

protocols; integrated with wide variety of tools45

4/17/2013

16

Security Services• Forms the underlying communication

medium for all the services

• Secure Authentication and Authorization• Single Sign-on

– User explicitly authenticates only once – then single sign-on works for all service requests

• Uniform Credentials• Example: GSI (Grid Security

Infrastructure)

46

• Authentication stops imposters• Examples of authentication:

– Username and password

– Passport

– ID card

– Public keys or certificates

– Fingerprint

Authentication means identifying that you are whom you claim to be

47

• Is this device allowed to access to this service?

• Read, write, execute permissions in Unix• Access conrol lists (ACLs) provide more

flexible control• Special “callouts” in the grid stack in job

and data management perform authorization checks.

Authorization controls what you are allowed to do.

48

4/17/2013

17

Job and resource management

49

Job Management Services provide a standard interface to remote resources

• Includes CPU, Storage and Bandwidth

• Globus component is Globus Resource Allocation Manager (GRAM)

• The primary Condor grid client component isCondor-G

• Other needs:– scheduling– monitoring– job migration– notification

50

GRAM provides a uniform interface to diverse resource scheduling systems.

User

Grid

VO

VO

VO

VO

Condor

PBS

LSF

UNIX fork()

GRAM

Site A

Site B

Site C

Site D

51

4/17/2013

18

GRAM: What is it?• Globus Resource Allocation Manager• Given a job specification:

– Create an environment for a job

– Stage files to and from the environment

– Submit a job to a local resource manager

– Monitor a job

– Send notifications of the job state change

– Stream a job’s stdout/err during execution

52

A “Local Resource Manager” is a batch system for running jobs across a computing cluster

• In GRAM• Examples:

– Condor– PBS– LSF– Sun Grid Engine

• Most systems allow you to access “fork”– Default behavior– It runs on the gatekeeper:

• A bad idea in general, but okay for testing

53

Managing your jobs• We need something more than just the basic

functionality of the globus job submission commands

• Some desired features– Job tracking– Submission of a set of inter-dependant jobs– Check-pointing and Job resubmission capability– Matchmaking for selecting appropriate resource for

executing the job• Options: Condor, PBS, LSF, …

54

4/17/2013

19

Grid Workflow

55

56

A typical workflow pattern in image analysis runs many filtering apps.

3a.h

align_warp/1

3a.i

3a.s.h

softmean/9

3a.s.i

3a.w

reslice/2

4a.h

align_warp/3

4a.i

4a.s.h 4a.s.i

4a.w

reslice/4

5a.h

align_warp/5

5a.i

5a.s.h 5a.s.i

5a.w

reslice/6

6a.h

align_warp/7

6a.i

6a.s.h 6a.s.i

6a.w

reslice/8

ref.h ref.i

atlas.h atlas.i

slicer/10 slicer/12 slicer/14

atlas_x.jpg

atlas_x.ppm

convert/11

atlas_y.jpg

atlas_y.ppm

convert/13

atlas_z.jpg

atlas_z.ppm

convert/15

Workflow courtesy James Dobson, Dartmouth Brain Imaging Center 56

Workflows can process vast datasets.• Many HEP and Astronomy experiments consist of:

– Large datasets as inputs (find datasets)– “Transformations” which work on the input datasets

(process)– The output datasets (store and publish)

• The emphasis is on the sharing of the large datasets

• Transformations are usually independent and can be parallelized. But they can vary greatly in duration.

Montage Workflow: ~1200 jobs, 7 levelsMosaic of M42 created on TeraGrid

= DataTransfer

= ComputeJob 57

4/17/2013

20

58

Virtual data model enables workflow to abstract grid details.

58

59

Grid Case Studies

• Earth System Grid• LIGO• TeraGrid

60

Earth System Grid

• Provide climate studies scientists with access to large datasets

• Data generated by computational models – requires massive computational power

• Most scientists work with subsets of the data

• Requires access to local copies of data

4/17/2013

21

61

ESG Infrastructure

• Archival storage systems and disk storage systems at several sites

• Storage resource managers and GridFTP servers to provide access to storage systems

• Metadata catalog services• Replica location services• Web portal user interface

62

Earth System Grid

63

Earth System Grid Interface

4/17/2013

22

64

Laser Interferometer Gravitational Wave Observatory (LIGO)

• Instruments at two sites to detect gravitational waves

• Each experiment run produces millions of files• Scientists at other sites want these datasets on

local storage• LIGO deploys RLS servers at each site to

register local mappings and collect info about mappings at other sites

65

Large Scale Data Replication for LIGO

• Goal: detection of gravitational waves• Three interferometers at two sites• Generate 1 TB of data daily• Need to replicate this data across 9 sites

to make it available to scientists• Scientists need to learn where data items

are, and how to access them

66

LIGO

4/17/2013

23

67

LIGO Solution

• Lightweight data replicator (LDR)• Uses parallel data streams, tunable TCP

windows, and tunable write/read buffers• Tracks where copies of specific files can

be found • Stores descriptive information (metadata)

in a database – Can select files based on description rather

than filename

68

TeraGrid

• NSF high-performance computing facility• Nine distributed sites, each with different

capability , e.g., computation power, archiving facilities, visualization software

• Applications may require more than one site

• Data sizes on the order of gigabytes or terabytes

69

TeraGrid

4/17/2013

24

70

TeraGrid

• Solution: Use GridFTP and RFT with front end command line tool (tgcp)

• Benefits of system:– Simple user interface

– High performance data transfer capability

– Ability to recover from both client and server software failures

– Extensible configuration

71

TGCP Details

• Idea: hide low level GridFTP commands from users

• Copy file smallfile.dat in a working directory to another system:tgcp smallfile.dat tg-login.sdsc.teragrid.org:/users/ux454332

• GridFTP command:globus-url-copy -p 8 -tcp-bs 1198372 \gsiftp://tg-gridftprr.uc.teragrid.org:2811/home/navarro/smallfile.dat \

gsiftp://tg-login.sdsc.teragrid.org:2811/users/ux454332/smallfile.dat

72

The reality

• We have spent a lot of time talking about “The Grid”

• There is “the Web” and “the Internet”• Is there a single Grid?

4/17/2013

25

73

The reality

• Many types of Grids exist• Private vs. public• Regional vs. Global• All-purpose vs. particular scientific

problem

74

Conclusion: Why Grids?

• New approaches to inquiry based on– Deep analysis of huge quantities of data

– Interdisciplinary collaboration

– Large-scale simulation and analysis

– Smart instrumentation

– Dynamically assemble the resources to tackle a new scale of problem

• Enabled by access to resources & services without regard for location & other barriers

74

75

Grids: Because Science Takes a Village …

• Teams organized around common goals– People, resource, software, data, instruments…

• With diverse membership & capabilities– Expertise in multiple areas required

• And geographic and political distribution– No location/organization possesses all required skills

and resources

• Must adapt as a function of the situation– Adjust membership, reallocate responsibilities,

renegotiate resources

75