Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 26 Jul 2010 12:04:34 +0200
From:      Ivan Voras <ivoras@freebsd.org>
To:        freebsd-stable@freebsd.org
Subject:   Re: gpart -b 34 versus gpart -b 1024
Message-ID:  <i2jmji$jjm$1@dough.gmane.org>
In-Reply-To: <4C4BB672.3090109@langille.org>
References:  <4C4BA50B.6050507@langille.org> <4C4BB672.3090109@langille.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On 25.7.2010 5:58, Dan Langille wrote:

>> -------Sequential Output-------- ---Sequential Input-- --Random--
>> -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks---
>> GB M/sec %CPU M/sec %CPU M/sec %CPU M/sec %CPU M/sec %CPU /sec %CPU

>> 5 110.6 80.5 115.3 15.1 60.9 8.5 68.8 46.2 326.7 15.3 469 1.4

>> 5 130.9 94.2 118.3 15.6 61.1 8.5 70.1 46.8 241.2 12.7 473 1.4

> 50 113.1 82.4 114.6 15.2 63.4 8.9 72.7 48.2 142.2  9.5 126 0.7

> 50 110.5 81.0 112.8 15.0 62.8 9.0 72.9 48.5 139.7  9.5 144 0.9

> Here, the results aren't much better either...  am I not aligning this
> partition correctly?  Missing something else?  Or... are they both 4K
> block aligned?

As others have said - your drives probably don't have the alignment
requiremnt, but your posts show in an excellent example why benchmarking
file systems is complicated and how easy it is to measure noise instead
of the real thing.

To measure real performance in your case, you would either need to
benchmark at a layer beneath the file system or with a simple file
system which does alwasy predictable io patterns. It's hard to do with
zfs with raidz - afaik even accessing the "raw" zvols translates into
complex IOs (they are COW).




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?i2jmji$jjm$1>