Voice Conference Meeting 16 February 2000
Ari gave a brief overview on how they have interfaced SRB into the LBL cache manager. Bing Zhu (SDSC) gave more details. If a user makes request from Wisconsin, LBL server will first send a request to the cache manager which will use SRB to locate the file and then SRB automatically transfers the file. Made tests on timeout (from cache manager), or if SRB driver talking to cache manager is not running, have made tests with files. Wisconsin will make test and put pressure on system, but want to separate the request from the transfer since canot assume all the requests (transfers) will be done, since need to allocate buffers first. HPSS resource manager may have their own policy, and what do you do with a requested but not transferred file, i.e. it is still in the cache, is there a priority to keep it in cache, but no guarantee. The SRB is working on parallel data transfers but that is not in production.
Ed May visited SDSC, and has installed the production version of SRB at ANL. Performance is not great (~200kbytes/sec), but this was expected. Using GSIFTP between ANL and LBL link (622Mbps/OC12) got about 18Mbytes/sec which has been consistent with 160-170Mbps measured by IPERF. Not sure why not getting closer to 622Mbps, probably could do better with parallel transfers. Routing thru ESnet (155Mbps/OC3) then got few Mbytes/sec. The above measures are with big windows but only one stream. Chuck Salisbury say 20-30Mbytes/sec last summer with a single stream. Ed had troubles getting the authentication installed, and also could not get the production client GSIFTP installed, so had to install a private version. It took him about 3 weeks to install. The level of support available from Globus is also interesting since their focus is more on research as opposed to production. This would be interesting if one is designing for a robust/stable system. We need someone from Globus to show what is needed to run GSIFTP, see http://www.globus.org/.
If we are focussing on the interfaces, then presumably we don't have to focus on the performance issues yet.
There are also questions about how Globus and SRB/MCAT seem to overlap/conflict in some areas.
There were at least 4 talks that referred to PPDG. There is a much greater understanding (since CHEP98) of how Objectivity works and there is less emphasis on this for between site access. Harvey says he felt there is a misconception of the volumes of data for LHC and RunII, the meeting said the difference was only a few times, but Harvey feels it is likely to be closer to 100.
See also Rough Notes on HENP Grid Meeting Padova, Italy Feb, 2000. Harvey reported that about 100 people showed up. Saturday was devoted to status reports. There has been a big change since last summer in Europe in that there is enthusiasm and things are optimistic for getting funding to support grids. There is an EU grid initiative, which mentions without much specificity that there should be coordination with the US. It is being spearheaded by CERN. The PPDG work is seen as being very important. The big issues is common infratrsucture and direction, e.g. how does one get the facilities to open themselves up to provide resources, another big issue is how to get security. The EU funding is expected to be tied closely to industry and getting funding depends on having shown one has delivered to industry in the past. There is a lot of effort/paperwork required to get the EU money, and it is by no means guaranteed taht any of the HENP related proposals will get funded. There are a number of separate national initiatives, this may make it difficult to build/deliver international grids. There will be a grid workshop at CERN on march 7, 2000.
Vicky raised the issue of the need for the people in the project to get togther physically. Stu will poll people on this, in particular to see if there is another meeting it can be coordinated with. Vicky is hoping we will be able to continue getting funding from DoE for the PPDG activities. She hopes she may be able to say something in a couple of weeks.