On 10/24/08 14:09, Somenath Bandyopadhyay wrote: > with intrsta I see only cpu#6 is getting interrupted... > > device | cpu0 %tim cpu1 %tim cpu2 %tim cpu3 %tim > cpu4 %tim cpu5 %tim cpu6 %tim > -------------+--------------------------------------------------------------------------------------------------------- > asy#0 | 0 0.0 0 0.0 0 0.0 0 0.0 > 0 0.0 0 0.0 0 0.0 > ata#0 | 0 0.0 0 0.0 0 0.0 0 0.0 > 0 0.0 0 0.0 0 0.0 > bnx#0 | 0 0.0 0 0.0 0 0.0 0 0.0 > 0 0.0 0 0.0 161 0.1 > bnx#1 | 0 0.0 0 0.0 0 0.0 0 0.0 > 0 0.0 0 0.0 161 0.0 > bnx#2 | 0 0.0 0 0.0 0 0.0 0 0.0 > 0 0.0 0 0.0 161 0.0 > bnx#3 | 0 0.0 0 0.0 0 0.0 0 0.0 > 0 0.0 0 0.0 161 0.0 > ehci#0 | 0 0.0 0 0.0 0 0.0 0 0.0 > 6 0.0 0 0.0 0 0.0 > mfi#0 | 0 0.0 0 0.0 0 0.0 0 0.0 > 0 0.0 0 0.0 0 0.0 > uhci#0 | 0 0.0 0 0.0 0 0.0 0 0.0 > 6 0.0 0 0.0 0 0.0 > uhci#1 | 0 0.0 0 0.0 0 0.0 0 0.0 > 0 0.0 1 0.0 0 0.0 > uhci#2 | 0 0.0 0 0.0 0 0.0 0 0.0 > 6 0.0 0 0.0 0 0.0 > > with mpstat (watch intr and ithr) shows all cpus getting interrupted but > interrupt threads running on cpu#6 making it >90% busy!
It would be interesting to see the output under load. I am not familar with bnx, so if the output above is under load, something doesn't match with the data below. The sum of 4x161 should be somewhat close to the number of interrupts in mpstat below. I also understand that crosscalls also generate interrupts, so that would explain why all CPUs have such high counts. Crosscalls suggest cache invalidation, so I do wonder whether psets for the user processes might help. Can't do that for NFS workload. Is the below output for NFS (since there is virtually no usr time)? You might want to tune via the bnx.conf file, looking at the tunables focusing on receive segments per interrupt. Also, maybe more buffers if you increase the segments per interrupt, so you don't run out. I noticed that the comments for the default values and the actual values don't match. If you follow the suggestions from others to place a service call, you may want to include that as well. Steffen PS. what type of system is this? > CPU minf mjf xcal intr ithr csw icsw migr smtx srw syscl usr sys wt idl > 0 0 0 30585 39533 239 2961 43 376 332 0 586 0 38 0 62 > 1 0 0 36700 37833 0 2685 28 363 260 0 673 1 39 0 60 > 2 0 0 59668 33989 101 1162 11 67 638 0 1062 1 49 0 50 > 3 0 0 63068 33893 0 1156 12 44 376 0 1143 0 50 0 50 > 4 0 0 28862 38099 3 564 4 23 269 0 529 0 37 0 63 > 5 0 0 20677 39201 1 411 4 20 207 0 377 1 30 0 69 > 6 0 0 3480 47606 5717 67 14 7 513 0 119 0 92 0 8 > 7 0 0 51324 35368 0 1076 3 42 277 0 901 0 42 0 58 > -- > This message posted from opensolaris.org > _______________________________________________ > networking-discuss mailing list > [email protected] _______________________________________________ networking-discuss mailing list [email protected]
