Comment 2 for bug 793437

I ran a test of my own devising, in which I wrote a trivial C program that calculates the square of a number using a quadratic algorithm, built it, and ran it under both the old and new kernels. Here are the results:

[2.6.35-28 kernel]
chad@siga-asus:~/temp/testing$ time ./a.out 20000
The square of 20000, calculated very slowly, is 400000000

real 0m1.045s
user 0m1.040s
sys 0m0.000s

[2.6.38-8 kernel]
chad@siga-asus:~/temp/testing$ time ./a.out 20000
The square of 20000, calculated very slowly, is 400000000

real 0m53.264s
user 0m49.130s
sys 0m3.100s

Since the "real" time is roughly the sum of "user" and "sys", this seems to be a confirmation that the CPU is the problem, rather than memory or I/O transfers. More interestingly, the "user" time dominating the "sys" time, which seems to indicate that it is not just system calls that are running at a glacial pace.