1. Grid

2. Introduction to Grid Computing  What is a Grid – an integrated advanced cyber infrastructure that delivers: oComputing capacity oData capacity oCommunication capacity  Why? There are many applications that are characterized as follows: oLarge varied distributed collaborations need to work together oNeed lots of cycles, storage (we are talking about teraflops, terabytes) oNeed to share results, codes, parameter files, …

3. Grid Motivation  Grid Computing was originally about extending scientific computing on single machines to distributed systems  Despite the improvement in raw computing power, storage capacity, communication it is difficult to keep up with the increased demand from the types of applications being developed.

4. Grid Motivation  Scientific Applications oAnalysis of large data volumes from different sources. oLots of computation needed to model an aspect of the natural world oOften requires substantially different types of computational resources oProjected data is measured in petabytes; Lots of storage

5. Grid Motivation  Astronomy oDigital sky surveys  Medical data oX-Ray, mammography data, etc. (many petabytes) oDigitizing patient records (ditto)  Molecular genomics and related disciplines oHuman Genome, other genome databases oProteomics (protein structure, activities, …) oProtein interactions, drug delivery  Virtual Population Laboratory (proposed) oSimulate likely spread of disease outbreaks  Brain scans (3-D, time dependent)  Climate studies

6. Grid Motivation  In the business world, companies want to integrate, manage and analyze large volumes of data oExample: An insurance company mines data from partner hospitals for fraud detection

7. Grid Motivation  Could buy additional machines  There is a lot of computing power that is unutilized or underutilized most of the time  How can applications take advantage of the multiple resources available in an effective manner?  A grid is intended for allowing the sharing, selection and aggregation of a wide variety of geographically dispersed resources owned by different organizations (virtual organizations)

8. Emergence of the Virtual Organization “ Resource sharing & coordinated problem solving in dynamic … virtual organizations” “The Anatomy of the Grid”, Foster, Kesselman, Tuecke, 2001

9. Other Distributed Infrastructures  Road, rail, telephones, power, banking, water, electrial  All started locally, then regionally, then nationally, and then internationally  Provide reliable relatively low cost access to a standardized service  Available to the masses

10. Electrical Power Grid  Single entity providing power  Relatively efficient, low cost, reliable  US Grid links 10K generators  Complex physical connections and trading mechanisms  Components heterogeneous and operated/owned by different companies  Consumers differ in amount of power they use, the quality of service they require, and the price they will pay  Economics important: grid driven by economic factors. Reserve capacities, trading power.  Politics important: success depended on regulatory, political and institutional developments as much as technical innovation  Control important: infrastructure for monitoring, management and control

11. Emergence of the Virtual Organization  Commonalities oNeed to discover and share resources oDo not necessarily trust all other participants oNot just about document exchange; Also about remote software, computers, data, sensors, etc; oResource sharing is conditional and the conditions are dynamic —Can only use resources for a limited class of problems or at certain times of the day.

12. What is a Grid Checklist (Foster)  Coordinates distributed resources using non-centralized control mechanisms. oA grid integrates and coordinates resources and users that live within different administrative domains —E.g.., different administrative units of the same company or different company oAddresses the issue of security, policy, payment, membership

13. What is a Grid Checklist (Foster)  Uses standard, open, general-purpose protocols and interfaces oA grid is built from multi-purpose protocols and interfaces that address such fundamental issues as authentication, authorization, resource discovery, and resource access.  Deliver nontrivial qualities of service oResources should be used in a coordinated fashion to deliver various quality of service oQuality of service is usually defined in metrics such as response time, throughput, availability, etc;

14. Grid vs. Internet?  We’ve had computers connected by networks for 20 years  The Grid brings additional notions oVirtual Organizations oInfrastructure to enable computation to be carried out across these —Authentication, monitoring, information, resource discovery, status, coordination, etc  Can I just plug my application into the Grid? oNo! Much work to do to get there!

15. Are these Grids?  Cluster Management Systems oExamples: Sun’s Sun Grid Engine,Platform’s Loadsharing facility oThese can be installed on a parallel computer or in a local area network oCan deliver a quality of service oEach may be an important component of a Grid, but by itself does not constitute a Grid

16. Are these Grids?  Multi-site scheduler oExample: Platform’s multicluster scheduler oYes: Not terribly sophisticated but it is a grid  Gnuetella oMaybe – Is it too specialized. oIs it open or is it a standard?  WWW  Foster’s checklist more clearly applies to large- scale Grid deployments: oData Grid: GriPhyN, PPDG, EU DataGrid, iVDGL, DataTAG, NASA’s Information Power Grid oTeraGrid: Used to link major US academic sites

17. Advantages of Grid Computing  Uses resources scattered across the world oAccess to more computing power oBetter access to data oUtilize unused cycles  Facilitates Virtual Organizations (VO) oGroups of organizations that use the Grid to share resources

18. DOE X-ray grand challenge: ANL, USC/ISI, NIST, U.Chicago tomographic reconstruction real-time collection wide-area dissemination desktop & VR clients with shared controls Advanced Photon Source Online Access to Scientific Instruments archival storage www.globus.org

19. Image courtesy Harvey Newman, Caltech Data Grids for High Energy Physics Tier2 Centre ~1 TIPS Online System Offline Processor Farm ~20 TIPS CERN Computer Centre FermiLab ~4 TIPS France Regional Centre Italy Regional Centre Germany Regional Centre Institute Institute ~0.25TIPS Physicist workstations ~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 Caltech ~1 TIPS ~622 Mbits/sec Tier 0 Tier 1 Tier 2 Tier 4 1 TIPS is approximately 25,000 SpecInt95 equivalents www.globus.org

20. NEES (Network for Earthquake Engineering Simulation) Collaboration U.Nevada Reno www.neesgrid.org

21.  Community = o1000s of home computer users oPhilanthropic computing vendor (Entropia) oResearch group (Scripps)  Common goal= advance AIDS research Home Computers Evaluate AIDS Drugs www.globus.org

22. SHARCNET is a high performance scientific computing project involving the University of Western Ontario, University of Guelph, McMaster University, the University of Windsor and Wilfred Laurier University. SHARCNET provides UWO researchers with world-class computing capabilities. As of November 2001, the computer cluster at the University of Western Ontario was the fastest computer at a Canadian University and the 12 th fastest in any University in North America. Cluster of Clusters or “Super Cluster” Ultra high speed fiber optic networking South Western Ontario Guelph Hamilton Waterloo London Windsor The University of Western Ontario McMaster University University of Guelph Wilfred Laurier University University of Windsor SHARCnet http://www.sharcnet.ca Sharcnet

23. Example Grids NSF PACI Grid NorduGrid

25. Ideal Grid-based Scientific Computation  User submits request through GUI oApplication oOperating System and other requirements oInput data  Grid finds and allocates resources to satisfy request  Grid monitors request processing oMoves job when resources fail or are too busy  Grid notifies user when results are available

26. Example  Assume a source file Main.F on machine A, an input file on machine B. Main.F is written using MPI, it will need around 4GB of core memory to run, it will take several hours to complete, and will produce a large output file.  What functionality is needed?

27. Issues  How to select a machine to run it on?  How to provide an executable which can run on that machine?  How to move the input file?  How to start the executable?  How to monitor the job? When does it start? When does it finish?  How to move the output file back?  What about security?  How do we know if it didn’t work and how it failed?

28. How to Select a Machine  What properties of a machine are we interested in? oWhat resources does my executable require? —4 GB memory, “several hours of compute time” —Enough diskspace for the output oWhat kind of environment do I need on the machine? —OS limitations? —MPI? (Which version?), Fortran? oWhat resources am I authorized to run on? oHow quickly will it run? oHow much will it cost/what is my allocation there? oHow to find all this information? What should the user provide?

29. More Complicated  What if the program might need to read in data kept on machine C while it is running?  What about distributing across processors on different machines?  What if I have a lot of interconnected programs?  How do I find the output file afterwards?  What if it doesn’t work?

30. Common Features Needed by Grid Systems  Resource registry is an information source that allows entities to publish and update information about the resource they wish to share Figure from Sean Norman’s reading course presentation

31. Common Features Needed by Grid Systems  Client is typically an agent acting on behalf of the user oAcquires resources requested by the user by consulting resource registries oSubmits an allocation request to the resource manager(s) responsible for the desired resources

32. Common Features Needed by Grid Systems  If request can be accommodated, resource manager(s) update status information for acquired resources in resource registries  Client then sends the appropriate executables and input data to the allocated resources and receives a reference to the execution in return

33. Common Features Needed by Grid Systems  Reference allows the client to monitor the execution of a job and inquire about its status  Client may also receive the results of the job once its execution is complete

34. Some Solutions  Middleware Toolkits: not all speak (or spoke) Globus: oCondor oGlobus Toolkit oLegion/Avaki oCondor (now Sun Grid Engine)  Higher Level Toolkits (build on Globus) oJavaCoG oGridPortal Toolkit, Grid Portal Development Toolkit (GPDK) oCondor-G oSGE