Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 6 Jul 2015 16:23:16 -0500
From:      Dave Duchscher <daved@nostrum.com>
To:        fs@freebsd.org
Subject:   ZFS system lockup
Message-ID:  <1BCFA515-BF3D-4E64-B826-BA475B13E770@nostrum.com>

next in thread | raw e-mail | index | archive | help
In the process of diagnosing an IO performance problem with our virtual =
environment, we ran into FreeBSD instances used in testing locking up =
and needing to be reset.  Moving to real hardware and running the same =
tests, we are able to reproduce the lockup.

We are testing using fio running a few read and write tests over and =
over again.  Watching via top, the system locks up and last update from =
top is reporting wired memory has taking all the memory (2G in the =
system, top shows1947M wired). ARC size at the time of the latest lockup =
was around 437M.  I can keep the system from locking if I reduce the =
maximum ARC size to 512M and wired memory floats around 1G. Setting =
maximum ARC 768M or higher and we get consistent lockups after running =
for a few hours.

What is using this wired memory?
Is there a way to keep wired memory under control with ZFS besides =
shrinking the ARC cache?
Is there any guidance on how much wired memory will be used for various =
ARC sizes?
Is 2G just too little memory to run ZFS?

We understand that the maximum ARC size will need to tuned in some cases =
but shrinking it down to 512M seems low.

This test hardware has a single 250G disk and 2G of RAM.  OS is FreeBSD =
10.1 Release.  Upgrading the system to stable and saw similar results.  =
Currently, the system is running 10.1 Release since that is what is used =
elsewhere.

We have seen a lockup on one of our database nodes which has 20G of RAM =
which we thought was caused by a SAN switch on our VM system.  Now we =
are not so sure.

--
Dave




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1BCFA515-BF3D-4E64-B826-BA475B13E770>