- Dteam Meeting (12/09/06) - Present: O. van der Aa P. Grongbech M. Leese (problems with vrvs) G. Cowan J. Jensen J. Coles D. Colling P. Gronbech S. Burke S. Traylen P. Strange PG: Problem having high throughput from Birmingham. Choose Oxford and the rate was very bad. Did some iperf tests and see a serious drop off and the suspicions is that it is due to the firewall installed in August. The plot from the net box show a drop at the time the firewall was installed. The physics firewall does not seems to be the problem GC: Summary of the tests for Scot grid. End of last week ed-durham tests. Durham has a limit of 200Mb/s and they hit it in both directions. Yesterday lanc-ed across Super Janet. Lancaster has a 100Mb/s on their production network. Rate was steady. JC: Why is Birmingham not being used as a reference site. PG: Almost 250Mb/s and it seems that there is a traffic shaping being put in place. It seems that wan people are putting shaping in place. JC: We should contact the local network people. PG: They usually answer that the network is underutilized and that it is not a problem to hammer it. Network monitoring: (Mark has problems via vrvs) ------------------------------------------------ JC: Is the webpage the updated version of the network monitoring page? ML: The pre-release sneak preview of the new version is at http://gridmon3.dl.ac.uk/gridmon/graph2.html OV: Imperial/UCL. Imperial given shipping address. Waiting for William to agree for UCL-CENTRAL DC: If rpm are given and Kostas reinstall the stuff as required then it is ok. DC: How much do we care about the T2-T2 rate. DC: The most monitoring we have the best, like I did to spot the oxford problem. JC: Gone trough the status of the sites: LT2 --- Brunel: Deployed - awaiting investigation into TCP performance Imperial: IP details received 11th September, awaiting dispatch of machine QMUL: Received on-site 8th September. Awaiting power and connection to network. RHUL: Deployed, although RHUL firewall currently prevents access to results database at Daresbury UCL: Waiting for IP details. NorthGrid --------- Daresbury: Deployed, although test box is currently not the same spec Lancaster: Machine on-site, although currently down - awaiting local investigation Potential issue with Lancs Web cache prevents download of gridmon software Liverpool: Deployed Manchester: Deployed Sheffield: Received on-site 9th June. Awaiting power and connection to network. ScotGrid -------- Durham: Deployed Edinburgh: Deployed, although machine currently down - awaiting local investigation Glasgow: Deployed. ICMP issues resolved 8th September SouthGrid --------- Birmingham: Deployed, although B'ham site firewall prohibits ICMP Have contacted B'ham Computing Services Bristol: Deployed Cambridge: Received on-site 17th July. Awaiting on-site electrical test Oxford: Deployed - awaiting investigation into TCP performance RAL tier 2: Deployed Comments about the web interface: A full mesh would be very useful ML: Probably not easy, probably a scaling problem. SB: Why is it a scaling problem? JC: Could be a useful tool if there is a complete mech. PG: This is to be used when there is a problem. PG: We can have a mesh for inter t2 test ? JC: We should move on since the connectivity with Mark is difficult. JC: Conclusion: Clearly useful but would be nice to have a full mesh Marco Visit: ----------- JC: - Marco is setting up the Australian Tier2. He visited a quarter of the sites and made a summary. The conclusions are in the url attached to the agenda. - Some sites are under resourced because they don't find the right people PG: Probably because when coming at Oxford there was a post open but nobody to fill it. JC: Situation for oxford solved. JC: For Scot Grid is the situation solved ? Graeme is doing the cluster installation and the coordination of ScotGrid. JC: At UCL is Austin new or he was in the activity. OV: Don't know. Alice and myself are giving him help to be sysadm of UCL-CENTRAL JC: Potential issues: Mixture of different hardware. Don't think we can make something about this. JC: Did someone buy woodcrest DC: Yes at Imperial we did because the cost was much more competitive per ksi2k then opterons. JC: recommendations point 1. JJ: We have that kind of model. PG: I am a bit surprised with that. JC: Point 2 DC: Difficult to take into account. JC: Point 3, template for tenders PG: Could be useful DC: It is such a small time we spend on this. The first one I wrote I took from Andrew. PG: It is useful to have for example of the T1 tender. They are a lot of little details that can be useful. DC: I think that is true but having the very detail is not an option since things changes ST: If anyone can have the tenders we can give it. DC: They are probably published in the European journal for tenders. But writing a right tender does not prevent from having problems. JC: Can probably link to it PG: Useful to have not to loose tender document. JC: send comments to Jeremy. AOB: David, Grieg, Olivier, Derek is going to the Technical Meeting. Series of talks at the GDB see the GDB TPM shifts: Problem with the takeover. The shift of yesterday is having problems. PG: Looking at the tickets quite often they are assigned by Valery and he is doing more then we do. SB: He is backup and should not be doing most of the work. SB: There is a savannah for putting request/bugs for TPM. PG: There was a ticket that of which the state could not be changed. Comments on the All Hands paper of SB: Any comments? Needs comments by the end of this week. Final Item: GridPP3 proposal was reviewed. Specific comment about the support of storage. A comment should be written. JC: What is our strategy. Is it to have two solutions or only one. What we need to document is that one solution is not an options. Basically DPM is ok for small sites and dCache for bigger once. JC: We need to explain why we want to be funded for two solutions. JJ: I have looked at storm and the conclusion is that since it requires a commercial file system it is not an option. JC: Need to document this. JC: Comment the mail about the storage units.