The U.S. Department of Energy's Energy Sciences Network (ESnet) continues to roll out its next-generation architecture on schedule with the March 14 completion of the Long Island Metropolitan Area Network, connecting Brookhaven National Laboratory (BNL) to the ESnet point of presence (POP) 60 miles away in New York City. The new design is built on a diversely routed dual 10 gigabits per second (Gbps) dense wave division multiplexed (DWDM) ring that connects the DOE Office of Science laboratory to both the ESnet IP core and the new Science Data Network (SDN) for high-throughput science data and collaboration.
This service upgrade provides BNL with a fully usable aggregate bandwidth of 20 Gbps, an eightfold increase over the lab's previous 2.4 Gbps link to Manhattan. This architecture is designed to be easily expanded to eight waves supplying 80 Gbps to the lab. BNL is the largest Tier 1 data distribution site for the ATLAS experiment at CERN and requires this next-generation network to support its ATLAS mission.
One of these two 10 Gbps lambdas in place at BNL will describe access the ESnet SDN and be provisioned primarily for dedicated circuits scaling to support the demands of petabyte data transfers, such as those required by ATLAS. The data will be generated by the ATLAS experiment to be conducted on the Large Hadron Collider (LHC) currently under construction at CERN. This data will be carried from CERN to New York City by USLHCnet the American LHC Network).
In preparation for the experiment going on line and to ensure the network and data centers can accommodate the flood of data, a series of data challenges with simulation data have been carried out. Following the completion of the recent fourth and final LHC Data Challenge prior to production, ESnet contacted BNL to find out how this new network is helping them support the mission of BNL within the ATLAS collaboration. ESnet's Mike O'Connor sat down with BNL's CIO Thomas Schlagel; Bruce Gibbard, the head of ATLAS computing at BNL; and Scott Bradley, manager of data networks and voice services.
Question: Tom, how important is the success of the ATLAS program at BNL to the mission of the laboratory and how will BNL benefit from its long-term success?
Tom Schlagel: The success of the ATLAS experiment and the Tier-I US-ATLAS center is extremely important to the mission of the laboratory in conducting world-class research. In addition to the science aspects, the ATLAS and CMS experiments at the LHC are at the forefront of global collaborative computing using Grid technologies. As has happened frequently in the past, the particle physics community is driving computing innovation which will have long-term consequences outside of the scientific research arena. BNL will benefit directly by participating in these activities. The hope is that the computing and networking capabilities put in place to support ATLAS, along with the experience gained by putting them in place, will be beneficial to other research programs in the future.
Question: Bruce, as the head the ATLAS Tier 1 Computing Facility at BNL, how would you describe the role BNL plays in the collaboration?
Bruce Gibbard: The ATLAS collaboration will process and distribute its physics data using a hierarchy of computing centers. The sole Tier 0 center is located at CERN in Switzerland, while the next level of 10 Tier 1 centers is globally distributed in Europe, Tiawan, Canada, and the U.S. The Tier 1 sites are each responsible for archiving an allotted portion of the “raw” detector data, as well as providing the compute cycles required to process, and reprocess as necessary, that portion to produced the reduced data sets required for detailed analysis.
BNL is the largest of the ATLAS Tier 1 centers and the only one in the U.S, and so is responsible for archiving and processing approximately 20 percent of the ATLAS raw data as well as serving the processed data out to individual ATLAS users and sites in the U.S. Once a new processing pass on the data is completed, based on the latest version of the reconstruction software and calibration data, the Tier 1 sites exchange copies of the portions they processed. These consist of ESD (Event Summary Data) and AOD (Analysis Object Data). This results in each Tier 1 receiving, archiving and then serving as a distribution point for the ATLAS ESD and AOD data sets.
The next level of the hierarchy, the Tier 2 centers located at universities and research institutions, is where the majority of the researchers are expected to work. For ATLAS in the U.S., there are currently three Tier 2 centers, and two more will be selected for funding this year. The Tier 2 sites draw on the data sets made available at the Tier 1 sites.
Question: With the tremendous amount of raw LHC data acquired, then processed into additional data sets, exchanged between Tier 1 sites and distributed to Tier 2 sites, resulting in a total data volume of 5 PB at BNL in 2008, is the new LI MAN up to the task?
Bruce Gibbard: Our wide area bandwidth projections indicate that ATLAS should be adequately provisioned for bandwidth through 2008 and we have gained substantial additional confidence from participation in the LHC Data Challenge 4 exercise. During this multi-week exercise, we were able to sustain an average transfer rate from CERN to our disk arrays of 191 Mbps (~1.5 Gbps) compared to a target rate of 200 Mbps. This was actually a higher average rate than that achieved by any other Tier 1 site. However, our ATLAS bandwidth projections do not account for bandwidth contention between ATLAS and other BNL projects, such as RHIC. I was pleased that during the data challenge exercise we did not have to take any special action to restrict ATLAS use of network resources. We did have some concerns going into it about the RHIC project moving very large data sets from BNL to the Riken Center in Japan during the data challenge, but thanks to the recent upgrade of our network to a fully duplex 10 Gbps path, it all just worked.
Question: Scott, as the manger of BNL's data networking and voice services, you find out pretty quickly when the network is not performing. Would you please describe how this new network is performing and a little bit about the process that put it in place.
Scott Bradley: The performance of the new Long Island Metropolitan Area Network has been outstanding and we're very pleased with it. It's a high performance network that was the product of an extremely successful coordinated effort between BNL and ESnet. BNL took advantage of the technical expertise, project management and financial backing of ESnet that enabled us to exploit the detailed knowledge we've acquired through our many years of dealing with the local carriers in the Long Island and New York metropolitan area. Working with ESnet in this way has saved countless man hours and budget dollars otherwise spent on WAN circuit and dark fiber leases we would have needed to reach New York and Chicago. BNL was in effect freed to concentrate effort on our core competencies. Did I mention how well we scored in (LHC) Service Challenge 4 (smile)? I believe that this partnership has become much stronger as a result of this process and as we move forward we will continue to rely on ESnet for connecting to LHC ATLAS and its global collaboration.
ESnet, funded primarily by the DOE Office of Science to support its science mission, is managed by Lawrence Berkeley National Laboratory. ESnet services connect researchers at national laboratories, universities and other research institutions around the world. For more information, go to www.es.net.
Source: Lawrence Berkeley National Laboratory