https://www.gridpp.ac.uk/wiki/Core_Grid_services
- The main focus is currently perfSonar
Target for this meeting:
- Target for end of June? Another 4 sites active on the dashboard?
- Survey of DRI deployment issues / followup needed
- What do we want to measure (matrix of tests)?
Update from Mark:
I have updated the following parts of the wiki:
https://www.gridpp.ac.uk/wiki/Protected_Site_networking#UKI-SCOTGRID-GLASGOW
https://www.gridpp.ac.uk/wiki/PerfSonarInstall#PerfSonar_Documentation_Links ( Added Documentation Links)
https://www.gridpp.ac.uk/wiki/Core_Grid_services
For the Core Services task the priority is now getting everyone installed.
This is difficult due to the issues of time for all the sites. However, what we need to test and monitor is fairly straight forward.
As many sites are bringing on 10 Gig connections we need to look at some basic tests such iperf or tcpnut file transfers, I prefer iperf.
To gain a rough understanding of the capabilities of each connection. This should be recorded in the GridPP wiki. It will need a new page.
Then we should fire up Perfsonar between the sites at a basic rate of 1 Gig to start with for a couple of weeks once we have installed as many sites as we think is reasonable I would say 7 - 9 would do.
This will give us a basic understanding of the inherent latency and bandwidth limitations between sites.
The tests should be staged between RAL and the tier-2s and then the tier-2s to tier-2s.
When this is done we can then test the 10 gig links. I am vary of bursts of 10 gig traffic on the network at Glasgow but if other sites want to run at this line rate for bandwidth testing then that is a local decision.
All positive and negative results should again be recorded on a new wiki page.
From here we can build up an idea of how well all the connected sites are performing.
Then as we add more sites we have a foundation to build against.
After we are happy that the UK cloud is working correctly we can look at testing intra cloud as discussed before.
I will send you a separate email which documents the process for rate capping Perfsonar-PS boxes so that 10 gig connected sites don't accidentally flood 1 Gig connected sites.
Target for next meeting: