Processor cross-call

[tsuba]

Hi there,

We have a Sun Fire 6800 with 8 cpus, and users are complaining about performance of their application. The only thing I found not normal was the "xcall" in mpstat.
CPU minf mjf xcal intr ithr csw icsw migr smtx srw syscl usr sys wt idl
0 291 35 2308 2037 1930 418 8 25 949 0 3834 4 16 0 80
1 52 42 2862 20 2 1872 16 28 121 0 465 1 12 0 87
2 387 186 17499 34 1 1383 31 44 106 0 4717 3 18 0 79
3 749 42 4577 27 4 708 13 43 120 0 1639 3 14 0 83
8 1078 42 2942 22 1 512 15 38 105 0 1671 3 11 0 86
9 406 2 1317 14 1 693 9 24 58 0 9532 11 7 0 82
10 730 13 2201 17 1 1232 14 40 193 0 1451 3 8 0 89
11 810 13 1758 948 930 984 13 38 165 0 2466 3 11 0 86
CPU minf mjf xcal intr ithr csw icsw migr smtx srw syscl usr sys wt idl
0 0 3 303 1190 1089 76 0 0 447 0 25 0 3 0 97
1 0 2 472 6 3 1032 0 0 51 0 12 0 5 0 95
2 0 127 17398 7 0 1032 3 4 55 0 1968 2 23 0 75
3 0 3 888 99 95 374 0 8 25 0 3289 3 5 0 92
8 0 2 584 2 0 400 0 3 30 0 0 0 33 1 66
9 0 0 659 6 0 341 3 3 18 0 4168 6 4 0 90
10 0 6 402 6 0 785 2 7 87 0 211 0 1 0 99
11 0 2 531 478 474 447 2 4 36 0 266 0 0 0 100

The "xcal" statistics have high values (too higher I think). Does anyone knows what kind of situation leads to this problem? I know that "Xcal" shoud be less than 200. The users application is SAS v 8.00

Thanks in advance

HTT

 

[jimbopalmer]

CPUs 0 (and 11) seem to be doing a LOT of hardware interupts (ithr)

this may mean all the accesory cards are plugged into the bus run by CPU 0, it may be you want to spred out the cards in your box to different buses.

or I could be misreading this or that only applied to older suns

I tried to remain child-like, all I acheived was childish.