Current-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: WAPBL on multicore, 1 (ONE!) process...

> I'm officially blind. I should go unsober and stay away from my
> mail account for today. Apologies. I'm still seeing quite a worse
> ratio though.

I think I need to summarize the results from the tests I did:

1) Plain "build -j 16" on a WAPBL FS:  2 hours 19 minutes
2) As 1) but with a "while true; do sync; sleep 1; done" loop
   running in parallel: 22 minutes
3) A plain "build -j 1" on the same WAPBL FS: 2 hours 53 minutes
4) As 1), but with simonb's patch and vfs.wapbl.flush_disk_cache=0
   28 minutes 30 seconds
5) As 4, but with the shell loop from 2): 21 minutes 10 seconds
6) As 1, on a non-WAPBL FS: 25 minutes
7) As 6, but with the shell loop from 2): 27 minutes 15 seconds

What you're seeing with WAPBL and without is probably comparable
to the difference between my test cases 1) and 2), or perhaps
more appropriately 1) and 6), that's a factor of x6.3 or x5.5.

The difference between 4) and 5) is still interesting, and what I
concentrated on with the lockstat data collection.


- Håvard

Home | Main Index | Thread Index | Old Index