The Open Science Grid and how it relates to PRAGMA
0 likes58 views
The Open Science Grid (OSG) provides distributed computing services to advance open science. It has evolved from requiring organizations to deploy software to now offering operated services with minimal setup. OSG's goal is to enable any research group to benefit from its resources. It provides compute, data, and software services through a federated network and aims to facilitate open science collaborations across institutions. OSG indicated it can support PRAGMA's open science efforts and hopes to increase engagement with communities in the Pacific Rim region.
The Open Science Grid and how it relates to PRAGMA
- 1. THE OPEN SCIENCE GRID (OSG) And how it relates to PRAGMA by Igor Sfiligoi, UCSD/SDSC (isfiligoi@sdsc.edu) PRAGMA 37 - SEPT 13TH, 2019
- 2. Premise – Not the Grid you were used to • It is my understanding Grid has a negative connotation in PRAGMA • While OSG has “Grid” in its name, I believe it is far from the “Grid” you experienced 15 years ago • Please give me a chance to convince you of that PRAGMA 37 - SEPT 13TH, 2019 2
- 3. OSG evolution • 10 years ago, OSG provided an integrated software stack that organizations had to deploy in order for their researchers to benefit from OSG. •We learned the hard way that this is too high a bar to jump across for most researchers and institutions. • Today, we offer to operate services to the groups we serve in order to minimize the threshold to entry for everybody. • Leveraging work performed by the likes of PRP/NRP-P/TNRP (e.g. Kubernetes) PRAGMA 37 - SEPT 13TH, 2019 3
- 4. OSG Aspirational Goal A campus Research IT Organization should not have to learn anything ”non-standard” in order to have their researchers benefit from OSG, or have their resources be available via OSG PRAGMA 37 - SEPT 13TH, 2019 4 Now …. Not there yet, but getting closer by the day…
- 5. OSG Mission Advancing Open Science through distributed High Throughput Computing PRAGMA 37 - SEPT 13TH, 2019 5
- 6. Open Science •All of open science irrespective of discipline •Include the maximum possible dynamic range of science, groups, and institutions • From individual undergraduates to international collaborations with thousands of members. • From small colleges, museums, zoos, to national scale centers of open science. PRAGMA 37 - SEPT 13TH, 2019 6
- 7. What Sciences use OSG? • Order here is from most active to least active. • Category is based on self-identification … i.e. people say what they do. PRAGMA 37 - SEPT 13TH, 2019 7
- 8. Not all users are the same • The individual researchers and small groups • The campus Research Support Organizations • Multi-institutional Science Teams • XENON, GlueX, SPT, Simons, and many many more • Collaborations between multiple campuses • The 4 “big science” projects: • US-ATLAS, US-CMS, LIGO, IceCube Used services change between different categories PRAGMA 37 - SEPT 13TH, 2019 8
- 9. OSG Mission Advancing Open Science through distributed High Throughput Computing PRAGMA 37 - SEPT 13TH, 2019 9
- 10. distributed High Throughput Computing DRIVING PRINCIPLES: • Maximize total compute, do not optimize for any single application • Allow for resources coming from anywhere in the world • Embrace heterogeneity • Federated policy control PRAGMA 37 - SEPT 13TH, 2019 10 (as opposed to e.g. HPC)
- 11. distributed High Throughput Computing DRIVING PRINCIPLES: • Maximize total compute, do not optimize for any single application • Allow for resources coming from anywhere in the world • Embrace heterogeneity • Federated policy control IMPLICATIONS: • Expect high network latencies • Expect varying walltime between identical runs • Job restarts are the norm PRAGMA 37 - SEPT 13TH, 2019 11 (Although throughput should be high)
- 12. Thriving in dHTC • The path to successful dHTC involves two aspects: • Separate a big computing problem in many individually schedulable small problems. • Minimize your requirements in order to maximize the raw capacity that you can effectively use. • The infrastructure will transparently take care of job restarts Ingenious Parallelism PRAGMA 37 - SEPT 13TH, 2019 12
- 13. dHTC is special •dHTC scales by definition perfectly. • When a researcher understands how to partition their workflow into many individually schedulable compute problems they can scale out with ease to seemingly arbitrary scales of computing. • When we integrate all IT resources at Universities, National Labs, and the commercial cloud, we arrive at a near infinite resource pool. PRAGMA 37 - SEPT 13TH, 2019 13
- 14. Stronger together • For OSG VO, more than half of the compute comes from smaller sites (and this does not include any of the big sciences, that have their own pools) 15k cores PRAGMA 37 - SEPT 13TH, 2019 14
- 15. OSG IN DETAIL PRAGMA 37 - SEPT 13TH, 2019 15
- 16. OSG Compute Services • OSG provides the means for operating a compute federation • Mostly in the form of Metascheduling Service • But also operates one access point itself for the smaller players PRAGMA 37 - SEPT 13TH, 2019 16
- 17. Federation i.e. Distributed control • OSG works on three simple principles: • Resource Owners determine policy of use • This means that all policy of use is set locally by the clusters that join the federation. • Resource Consumers specify the types of resources they are willing to use. • How much RAM? How many cores per node? … • OSG submits HTCondor batch system as (pilot) payload into all local batch systems that match requirements, creating an overlay dHTC system • Jobs are submitted locally, queue centrally, and execute anywhere that matches requirements after resource becomes available. PRAGMA 37 - SEPT 13TH, 2019 17
- 18. OSG Data Federation • OSG operates a content delivery network (CDN) • Based on XRootD caching technology • Data origins operated by Science communities • Data transparently published in our federated namespace. • Caches hide data access latencies and reduce network traffic. PRAGMA 37 - SEPT 13TH, 2019 18
- 19. OSG Software Distribution •OSG can host and distribute community software on CVMFS • A CDN at heart, but based on Apache Squid technology • CVMFS provides a Read-Only POSIX interface •OSG encourages the use of containerization • Encourages participating sites to provide Singularity • Which is optimized for use by unprivileged users (unlike Docker) • Containers can be distributed through CVMFS (for performance) PRAGMA 37 - SEPT 13TH, 2019 19
- 20. Facilitation Services • Proactive, personalized guidance and support for: • Institution-specific submit points • Sharing institutional resources via OSG • Data federation across OSG sites • Individual researchers using the OSG submit infrastructure • We also offer • Local workshops • OSG-hosted education and training • Learning from the OSG Facilitation Community PRAGMA 37 - SEPT 13TH, 2019 20
- 21. OSG AND PRAGMA PRAGMA 37 - SEPT 13TH, 2019 21
- 22. OSG AND PRAGMA • OSG would be happy to facilitate any Open Science endeavor that has at least some of its members in the US • We are a NSF-funded project • All OSG services are at your disposal • Compute federation • Data federation • Software distribution • Facilitation PRAGMA 37 - SEPT 13TH, 2019 22
- 23. OSG and International Setups •All our major users are international • CMS, ATLAS, LIGO, IceCube, XENON, GlueX •But, alas, we have been mostly collaborating with Europe •PRAGMA could help us engage more Science communities in the Pacific Rim PRAGMA 37 - SEPT 13TH, 2019 23
- 24. Summary and Conclusion • OSG’s objective is to “Advance Open Science through distributed High Throughput Computing” • We are committed to help any Open Science community that has at least some members in the USA – Nice overlap with PRAGMA • OSG can offer a wide variety of services • Both in terms of technology and facilitation • We could use some help getting more involved in the Pacific Rim • Contact us at: help@opensciencegrid.org PRAGMA 37 - SEPT 13TH, 2019 24