Hi Zoltan, We too have seen poor performance on random access DISK volumes where multiple volumes are on the same underlying RAID set. A few things I've noticed:
* TSM is ignorant of the underlying filesystem structure. This means it will write new data and migrate/backup old data from the same filesystems at the same time, ignoring the fact that there's eligible data to migrate/backup on idle filesystems. You can observe this behavior on Linux using iostat. * Performance gets extraordinarily bad when the DISK pool fills up. You can easily get into a death spiral, where you can't migrate fast enough to keep from shoe-shining your tape drives to accomodate new data. The problem is when the pool itself fills up, not the underlying filesystem, so leaving headroom in the filesystem doesn't help you at all. My guess is TSM is having to try to span multiple DISK volumes when the pool fills up, which is pretty intensive. We're going to be replacing our disk spool in the next couple weeks. Rather than having small RAID-5 sets, we'll probably just have jumbo RAID-60 sets and just have a couple DISK volumes per pool. -- -- Skylar Thompson ([email protected]) -- Genome Sciences Department, System Administrator -- Foege Building S046, (206)-685-7354 -- University of Washington School of Medicine
