Subject: kernbench results
Newsgroups: gmane.linux.kernel.ck
Date: 2006-04-07 11:48:44 GMT (3 years, 12 weeks, 4 days, 10 hours and 22 minutes ago)
On Fri, Apr 07 2006, Con Kolivas wrote: > Just uploaded kernbench v0.40 to kernel.org which does the logging > automatically. kernel.org should sync up soon and I'll make a formal announce > of the new kernbench release later. Grab that one instead. Ok, so I did a run on both 2.6.17-rc1 and 2.6.17-rc1 + staircase. The hardware used was an Intel Tiger platform. It has 4 1.2GHz itanium2 CPUs with 32/256/6144KiB of l1/l2/l3 cache. The box has 4GB of RAM. Kernel source was hosted on a dedicated SCSI drive. It used gcc-3.3.3. Compile speed isn't very fast with gcc-3.3.3 on IPF. The 2.6.16.1 kernel was used as the source tree. I changed the numruns to 2 instead of 5 to get some quicker results, the initial run was still going after 50 minutes and I just don't have 2x1 hour to spend on that box there today! Sorry. Additionally the run that claims to use -j3 seems to show pretty much an exact 1.0 load and the runtime seems to indicate that is the case as well. I didn't look further, it may be kernbench and it may be the kernel build system. Not sure -j is ever a good idea btw, even on this box it spirals towards nothingness. -jLARGENUMBER is probably better. A quick look at the numbers show they are basically equal, so I'd say no regressions for this particular test. -- Jens Axboe
4 cpus found Cleaning source tree... Caching kernel source in ram... Making oldconfig... Half load is 2 jobs, changing to 3 as a kernel compile won't guarantee 2 jobs Kernel 2.6.17-rc1-g17111f0c-dirty Performing 2 runs of make -j 3 make -j 16 All data logged to kernbench.log Warmup run... Half load -j 3 run number 1... Half load -j 3 run number 2... Average Half load -j 3 Run (std deviation): Elapsed Time 1023.43 (0.106066) User Time 1012.84 (0.106066) System Time 25.495 (0.0353553) Percent CPU 101 (0) Context Switches 9469.5 (85.5599) Sleeps 32830 (0) Optimal load -j 16 run number 1... Optimal load -j 16 run number 2... Average Optimal load -j 16 Run (std deviation): Elapsed Time 282.935 (1.29401) User Time 1020.24 (8.55082) System Time 26.4275 (1.0839) Percent CPU 236.75 (156.755) Context Switches 20033.8 (12198.8) Sleeps 32508 (371.817)
4 cpus found Cleaning source tree... Caching kernel source in ram... Making oldconfig... Half load is 2 jobs, changing to 3 as a kernel compile won't guarantee 2 jobs Kernel 2.6.17-rc1-g17111f0c-dirty Performing 2 runs of make -j 3 make -j 16 All data logged to kernbench.log Warmup run... Half load -j 3 run number 1... Half load -j 3 run number 2... Average Half load -j 3 Run (std deviation): Elapsed Time 1022.61 (0.219203) User Time 1012.3 (0.113137) System Time 25.225 (0.00707107) Percent CPU 101 (0) Context Switches 9488 (84.8528) Sleeps 32834.5 (0.707107) Optimal load -j 16 run number 1... Optimal load -j 16 run number 2... Average Optimal load -j 16 Run (std deviation): Elapsed Time 281.835 (0.13435) User Time 1019.45 (8.25669) System Time 26.3175 (1.26194) Percent CPU 237.25 (157.328) Context Switches 19961 (12095.8) Sleeps 32531 (502.172)
4 cpus found Cleaning source tree... Caching kernel source in ram... Making oldconfig... Half load is 2 jobs, changing to 3 as a kernel compile won't guarantee 2 jobs Kernel 2.6.17-rc1-g17111f0c-dirty Performing 2 runs of make -j 3 make -j 16 All data logged to kernbench.log Warmup run... Half load -j 3 run number 1... Half load -j 3 run number 2... Average Half load -j 3 Run (std deviation): Elapsed Time 1022.61 (0.219203) User Time 1012.3 (0.113137) System Time 25.225 (0.00707107) Percent CPU 101 (0) Context Switches 9488 (84.8528) Sleeps 32834.5 (0.707107) Optimal load -j 16 run number 1... Optimal load -j 16 run number 2... Average Optimal load -j 16 Run (std deviation): Elapsed Time 281.835 (0.13435) User Time 1019.45 (8.25669) System Time 26.3175 (1.26194) Percent CPU 237.25 (157.328) Context Switches 19961 (12095.8) Sleeps 32531 (502.172)