Re: kern/49186: Network hangs when using write cache in amr(4) RAID controller

[David Holland <dholland-bugs%netbsd.org@localhost>]

The following reply was made to PR kern/49186; it has been noted by GNATS.

From: David Holland <dholland-bugs%netbsd.org@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: 
Subject: Re: kern/49186: Network hangs when using write cache in amr(4) RAID
 controller
Date: Thu, 11 Sep 2014 18:35:57 +0000

 On Mon, Sep 08, 2014 at 05:00:01PM +0000, tih%hamartun.priv.no@localhost wrote:
  > I'm running NetBSD/amd64-current on a Dell 2850 with an AMI PERC 4e/Di
  > RAID controller with 256MB of RAM, and three logical volumes, each
  > comprising two mirrored disks.  The above dmesg output shows three
  > network interfaces, but only wm0 is in use.  It also shows a degraded
  > logical disk (ld2), but the problem was present before that happened.
  > That particular volume is, in any case, mostly idle.
  > 
  > When enabling background writes in the cache configuration, heavy disk
  > activity leads to random network hangups, where all network activity
  > on the system will pause for anything from a few seconds to a couple
  > of minutes.
 
 It seems weird that a disk change should provoke network problems.
 (Is this machine an nfs server?)
 
 The only thing that comes to mind for me is that if you've been
 tinkering with the number of mbufs or similar config settings, try
 reverting that temporarily. I remember someone recently had a problem
 running out of mbufs that turned out to be caused by long-forgotten
 tinkering along those lines that skewed the space allocation in the
 kernel heap.
 
 I have no idea why background writes in the disk controller would
 provoke the situation, but maybe it's just because the machine runs
 faster that way and chews up its heap more readily.
 
 -- 
 David A. Holland
 dholland%netbsd.org@localhost