We have a Linux cluster (half dual Intel, half Alpha). In the past we have
encountered kernels that have unacceptably slow networking. I just
upgraded to RH 6.1 on several node machines and then further upgraded to
2.2.14 on some frontend machines (2 Intel, 1 Alpha). Some testing has
determined that the RawHide 2.2.14 kernel rpms appear to suffer the
slow-network bug, whereas the kernels shipped with RH 6.1 do not.
Specifically, shipping a ~20MB file with time rcp results in
2.2.14 Intel -> anywhere
0.010u 0.330s 1:12.63 0.4% 0+0k 0+0io 147pf+0w (typical)
Worst case: not sure yet as I have one rcp still hung after about 25
2.2.14 Alpha -> 2.2.13 Alpha took about 7 sec wallclock
2.2.13 Alpha -> same took
0.007u 0.541s 0:02.67 20.2% 0+0k 0+0io 2569pf+0w
Finally, 2.2.12 Intel -> same took about 4 sec wallclock
I'm going to back out 2.2.14 from those machines on which it was installed.
I have read that 2.2.12 and 2.2.13 have bugs (our intels are SMP) but we
will just see whether we can get by. Networking is critical to us.
No response, closing bug.
If you want to re-open it, please try the 2.2.16-3 errata kernel