The power of sharing in OSG

~Chander Sehgal, OSG project manager, with Greg Moore

The Open Science Grid (OSG) is a sharing ecosystem. It doesn’t “own” any computing resources. Members of the OSG consortium connect their resources to perform computing for their experiments — and also make available resources not currently in use for other OSG members’ experiments. The shared compute cycles are referred to as opportunistic.

These opportunistic cycles are helping US researchers accomplish science. The OSG plays a role, to be sure. But more importantly, the virtual organizations (VOs) that own the sites with computing resources are key to this sharing success. Researchers are enabling other researchers. This sharing culture enables science in new ways. The sites that provide opportunistic access are the unsung heroes in this success, and they deserve a big thank you for sharing around 80 million computing hours to enable US research.

Back in July 2013, we highlighted Snowmass, an example of a research group that doesn’t have its own computing resources. Dr. Meenakshi Narain gave a glimpse into the roles of the OSG and the group. Professor Narain is co-principal investigator of Brown University’s experimental particle physics group, which played a prominent role in the Higgs boson discovery. She is also one of the conveners of the Snowmass New Particles group, coordinator of the Fermilab LHC Physics Center (LPC), and lead for the OSG Snowmass group (which she founded).

Dr. Narain noted that the OSG was creating easier ways for various teams to collaborate and share simulated data. “The OSG brings the community together in a seamless way and helps us forge collaborations,” she said. Instead of duplicating each other’s work, researchers can compare data and models and share common tools. These collaborative efforts will help influence the design of future detectors at hadron colliders. Without the OSG’s culture of sharing, the Snowmass group’s critical role in the discovery of the Higgs boson might not have been possible.

This sharing culture enables peak management among the VOs. Anything not in use is made available to other researchers in the US. In total about 80 million hours of opportunistic cycles became available in the last 12 months; the major VOs who benefited from this are shown in Figure 1.Wall hours


Figure 1: represents wall hours by VO for a 52 week period- totalling 80,758,104 hours.

Hours spent on jobs


Figure 2: shows the weekly availability of cycles via opportunistic access, which averages around 1.5 million hours per week.

One of the VOs that benefited the most from these opportunistic resources is the OSG VO; it doesn’t own any computing resources and was created with one purpose in mind: Scavenge what is currently unused in the OSG ecosystem, and make it available for other US researchers to get their work done. But this sharing is really a team sport in OSG. Other regional teams such as GLOW, Engage, HCC, and other VOs are enabling a broad set of US researchers. Across the OSG, we have supported over 50 researchers in the last year, and this continues to grow.

Researchers can get complex scientific computation done in a much shorter timeframe because of the large pool of resources that the OSG makes available. For example, we highlighted work being done at the University of Wisconsin–Madison’s de Pablo Research Group, which studies the thermodynamic properties of materials such as DNA and proteins at the nanoscale. The group’s tools are all computational. OSG computing power helps them capture the details of how DNA molecules interact with histone proteins. They can run in excess of 3,000 simulations simultaneously, something they cannot do in their lab — and they can also break very long simulations into trivially parallel simulations that don’t need to communicate with each other. As a result, their productivity has skyrocketed.

Opportunistic hours on the OSG in the last year amounted to a little more than 80 million, or about 1.5 million per week. If buying CPU time costs 5 cents per hour, that’s about $4 million. Aside from the monetary value, this sharing makes it possible for researchers to get their work done on compressed intervals that would be impossible without access to such a resource. The only “cost” for researchers is acknowledging and citing the OSG, which is supported by the National Science Foundation and the Department of Energy’s Office of Science.

In a grid with over 100 sites, there is a good chance that there will be unused capacity at some sites at any point in time. The challenge is detecting availability in a timely fashion and matching it with the needs of researchers. OSG is currently working to improve our ability to identify and harvest these opportunistic resources. We are also increasing our efforts to reach out to US campuses to identify more researchers who can benefit from access to distributed high-throughput computing (DHTC) resources.