Subject: Re: Some kernel profiling on a 5000/133 with -current
To: Chuck Silvers <chuq@chuq.com>
From: Ken Wellsch <kwellsch@tampabay.rr.com>
List: port-pmax
Date: 08/30/2001 18:25:26
Well, I tried a quick head-to-head and just as you predicted, the change
has basically no effect. The calls to memcpy() clearly are reduced, but
without any significent effect. Ah well... sigh. I'll give up my novice
efforts.
-------Sequential Output-------- ---Sequential Input-- --Random--
-Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU
-current 128 533 94.9 1195 86.8 857 60.9 670 94.4 2027 65.8 170.0 52.1
new 128 530 94.3 1180 87.2 862 61.4 672 94.5 1997 65.0 161.3 49.8
Flat profile -current:
% cumulative self self total
time seconds seconds calls us/call us/call name
26.69 135.87 135.87 _splset_noprof
17.95 227.25 91.38 712374 128.28 128.28 memcpy
12.19 289.27 62.02 5505 11266.75 11266.75 mips_idle
5.48 317.15 27.88 _mcount
4.78 341.46 24.31 __mcount
2.19 352.61 11.15 813688 13.71 13.72 lockmgr
1.70 361.26 8.65 202244 42.76 42.76 mips1_FlushDCache
1.55 369.13 7.87 438844 17.94 26.48 pmap_enter
1.52 376.88 7.75 3821828 2.03 5.84 st_scsibus_cmprss
1.45 384.26 7.38 486170 15.18 15.18 memset
1.39 391.33 7.07 168455 41.97 207.94 genfs_getpages
1.32 398.05 6.73 20919 321.55 441.51 uvn_attach
Flat profile new:
% cumulative self self total
time seconds seconds calls us/call us/call name
26.27 134.29 134.29 _splset_noprof
18.30 227.84 93.55 416863 224.42 224.42 memcpy
11.97 289.04 61.20 5464 11201.16 11201.16 mips_idle
5.45 316.88 27.84 _mcount
4.76 341.20 24.32 __mcount
2.17 352.31 11.11 818332 13.58 13.59 lockmgr
1.84 361.70 9.39 202034 46.48 46.48 mips1_FlushDCache
1.53 369.52 7.82 3832726 2.04 6.04 st_scsibus_cmprss
1.51 377.22 7.70 439454 17.53 28.34 pmap_enter
1.47 384.74 7.52 487057 15.43 15.43 memset
1.46 392.18 7.45 168971 44.06 427.19 genfs_getpages
1.38 399.26 7.07 21554 328.21 460.65 uvn_attach