 |
 |
Today Bill and I attempted to start 1092 xrootd data servers and see if they could self-cluster and deliver data. I would call the test a success eventhough only 900 nodes were able to successfully come up. Which is excellent for this test. I traced the failures of 192 nodes to testing turds left behind within the last year. Specifically, clusters were started by various people on certain bronco's and tori's which caused files to be created with the wrong ownership; preventing the olbd from comming up as the files were either inaccessible or not overwritable due to permission problems. I will be looking to see how we can avoid that problem in the next release. Once we overcame some lsf scalability issues (we used lsrun to start the servers); the test went smoothly and 900 nodes were able to reach a stable configuration in about 56 seconds. I will be try to cluster all 1092 nodes next week. Andy