Eduard Bloch <[EMAIL PROTECTED]> writes: >> # time bzip2 -9k linux-2.6.0-test5.tar >> >> real 2m40.974s >> user 2m33.679s > ... >> user 2m49.316s > > Even then, it's about only 10 percent. Let's compare them with vanilla > kernel, optimised for P4:
What are you trying to measure here? If you want to benchmark the kernel, use something that spends a significant amount of time in the kernel. Programs like bzip2 spend almost all their time in userspace. Unless the kernel significantly changes algorithms for things like swapout, your benchmark says more about random variability than kernel performance. There are many meaningful kernel benchmarks: bonnie, contest, and many others. Those are the programs that will tell you how well the kernel performs. Michael