> We've seen a lot of semiscientific benchmark numbers thrown around
> lately.  I happen to believe that the only benchmarks that matter
> are properly written microbenchmarks whose correlation to
> architectural features is well understood, and your own application.
> And of course benchmarks where you don't normalize everything you
> can about the system under test are meaningless.

> Microbenchmarks are boring.  My application is compiling operating
> systems, because I do it many times a day for my job.
[skipped]

It whould be great to see the same tests for NetBSD's LFS file system
on the same hardware.  That is to test ffs vs. lfs for this particular
task.

-- 
Best regards, Aleksey Cheusov.