Linux 2.4.19-pre8-jam2
Quoted 6 lines
Quoted 4 lines
Quoted 2 lines
- Re-introduction of
wake_up_syncto make pipes run fast again. No idea about this is useful or not, that is the point, to test it (Randy ?)2.5 kernels <= 2.5.15 aren't completing umount on the 4 way Xeon.
Is umount not completing somehow due to the lack of
wake_up_sync???
The umount issue is unrelated to wake_up_sync. Could be a scsi driver issue, as 2.5.x has been fine on my IDE system. Before the benchmarks, I mkfs some filesystems and mount and umount them. Haven't been able to get past that point with 2.5.x yet.
Quoted 4 lines
Fwiw, I am not sold that reintroducing
wake_up_syncis worth it. The benchmark is synthetic and could very well not represent the general case in which the load balancer is capable of handling the scenario without the hackery of an explicit sync option.
That could well be. The kernel build test which applies patches and compiles with -pipe is as close as I have to a general pipe test. The kernel build test has a low variability (~2%) between any of the kernels tested on the 4 way box.
It's worth testing wake_up_sync though. Occasionally there's a surprise.
Signature · 1 line
Randy Hron