Date: Fri, 26 Jan 2001 12:49:58 +0300 From: "Artem Koutchine" <matrix@ipform.ru> To: <questions@FreeBSD.ORG> Subject: SoftUpdates + fs otions find benchmarks are here Message-ID: <007301c0877d$5b29c620$0c00a8c0@ipform.ru>
next in thread | raw e-mail | index | archive | help
Hi!
I've done some bechmarks on softupdates with different filesystem
options. Benchmark is a very simple one:
time find /usr
/usr is 7241594 1K block long, has
HARDWARE:
PIII 550Mhz, 128Mb RAM,
IDE QUANTUM FIREBALLlct10 10 9787Mb
MASTER UDMA-33
Interupts never went above 2%, system was never above 4%.
Here is what i've got (time ascending):
noatime+su
08m54.222s
noatime
08m55.301s
noatime+async
08m57.124s
noatime+async+su
08m58.251s
async+su
12m3.157s
su
12m5.452s
no options:
12m14.341s
async
12m23.763s
1% of worst time = 7 seconds.
Difference smaller than 3% should be accounted for
testing flaws.
I realize that async with atime is useless, since nothing
is written in this benchmark. Also, async should be useless
in this benchmark anyway. And using async+su together
is at least weird, because they do pretty much the
same job.
I;ve read the list and know that using atime+su used to
cause kernel panics but no more. So, it seems to be
fine (theoretically). Does anybody run it like this?
Weird result: no options at all seems to be faster then
async option along. However the differene is within 3%,
so we can say they perform the same. Does it mean
that async does not apply to updating atime?
Overall result: if you going to have a system which
does a lot of small read on a lot of files
(usual suspects: loaded web servers) using noatime
will save the day, all other option give negligible
win. I have not run any special test, but it seems that
mysql can gane some speed with noatime too. Any
ideas?
This is not a strict test, nor it is a complete one.
If you have some additional data or comment, please share it in
the list
I am planning to run some REAL fs benchmarks with different options
later.
Best Regards,
Artem
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?007301c0877d$5b29c620$0c00a8c0>