I'm currently moving everything from server a to server b. Speed is not impressing, these are times for the whole vos move operation (clone, forward, delete):
------ Speed was 599.818578741343 kbytes/sec (~6677 seconds) ------ Speed was 15243.7196110211 kbytes/sec (~ 590 seconds) and so on from my logfile ------ Speed was 10550.5408970976 kbytes/sec ------ Speed was 3802.03041825095 kbytes/sec ------ Speed was 6751.37335526316 kbytes/sec ------ Speed was 9946.13157894737 kbytes/sec All these volumes are > 3GB, so they are big enough to cancel out any start-stop effects. I looked at the first two volumes: The 599KB/sec volume has 4GB in 72441 files and the 15MB/sec volume has 9GB in 696 files. I think I see a pattern here. Calculating backwards I get 10 files per second for the volume with many files and about 1 file per second for the volume with few files. 15MB/sec is probably dominated by data move time and 10 files per second by file create and delete times. 15MB/sec is not good but 10 files per second is just horrible. The vos move contains different operations. One that seems to take very long time is the delete of the original volume. Is there any reason that the caller has to wait for that delete to complete? The files are allready copied and the VLDB is updated. So why not just put the old volume in the "never be able to come back from the dead state", start the cleanup and return the call? In that way my next vos move would start sooner and some concurrecy could be exploited. And yes, the hardware (linux-xfs on 5 SW raided SCSI disks, gigabit an 2 EMT64 CPUs) should be fast enough to be able to give me more than 599kbytes/sec. Harald. PS: 537055754 RW 3983995 K used 72441 files On-line 537064626 RW 9405386 K used 696 files On-line _______________________________________________ OpenAFS-devel mailing list [email protected] https://lists.openafs.org/mailman/listinfo/openafs-devel
