mailing list archives
Re: "IO wait chains" in Linux??
From: coderman <coderman () gmail com>
Date: Mon, 7 Feb 2011 16:49:05 -0800
On Mon, Feb 7, 2011 at 4:38 PM, Cal Leeming [Simplicity Media Ltd]
<cal.leeming () simplicitymedialtd co uk> wrote:
... IO wait / deadlocking was the bane of my existence for the entire
of 2010 whilst attempting to try out technologies such as DRBD, XFS, OCFS2
in a HA environment
in this case, there are a suite of tools for the kit. general idiom is
setup measurement and load injection, then executive to define peak /
best capacity configuration. (hah, so simple!)
iozone for disk / storage subsystem measurements. ex.:
iozone -e -i0 -i1 -i2 -i8 -+n -r 16k -s32g -t1
iozone -e -i0 -i1 -i2 -i8 -+n -r 16k -s32g -t2
iozone -e -i0 -i1 -i2 -i8 -+n -r 16k -s32g -t4
iozone -e -i0 -i1 -i2 -i8 -+n -r 16k -s32g -t8
iozone -e -i0 -i1 -i2 -i8 -+n -r 16k -s32g -t...
for memory use ramdisk based block, or native initramfs, or other
memory backed loops and paging.
oprofile / kprobes /sysprof help with measurement rather than load injection.
http://oprofile.sourceforge.net/examples/ - Introduction to KProbes
network load injection / measurement is another beast, perhaps someone
else can fill in that huge expanse of tangents.
Full-Disclosure - We believe in it.
Hosted and sponsored by Secunia - http://secunia.com/