Date: Tue, 20 Apr 2010 14:29:13 +0200 From: Attilio Rao <attilio@freebsd.org> To: David Ehrmann <ehrmann@gmail.com> Cc: freebsd-current@freebsd.org Subject: Re: Strange disk problem Message-ID: <t2o3bbf2fe11004200529h45cf209et731a194794aae7e1@mail.gmail.com> In-Reply-To: <4BCD5049.8030408@gmail.com> References: <4BCD5049.8030408@gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
2010/4/20 David Ehrmann <ehrmann@gmail.com>: > Initially, I noticed a problem where reading a file on this machine seeme= d > to stop--something like a video would just stop playing. =C2=A0At first, = I > thought it was the machine, but a new motherboard, CPU, and RAM later, th= e > problem persists. =C2=A0The network card uses a different chipset, too. > > The files are on zfs, but scrubs are fine, and zpool status lists no erro= rs > of any kind. =C2=A0Trying to reproduce the problem, I set up a script tha= t > reading a random 1M block every 60 seconds off the drive backing zfs. > =C2=A0That's when I noticed something: one disk seems to be causing the p= roblems. > =C2=A0I logged the dd times, and some of them were huge--more than a minu= te. =C2=A0The > times on the other disk in the mirrored vdev were low. > > I've only seen the problem when I have a vm's disk image hosted on the > machine. =C2=A0That said, the network interface is configured at 100mbps,= so > there's no reason for that to saturate the disk's throughput. =C2=A0Top r= eports > that almost 20% of the CPU is going towards interrupts. =C2=A0I can read = a file > off the zfs pool at over 50MB/s, so that shouldn't be a problem. =C2=A0On= e thing > I'm wondering is why the disk read doesn't timeout quickly? =C2=A0At leas= t that > way zfs could try to use the other drive in the mirrored vdev. > > Any ideas? =C2=A0One thing I should try is switching the drive, see if th= e > problem follows the disk or stays with the lowest /dev/adX device. =C2=A0= I'm > using geli, but the read problems happen with both /dev/adX AND > /dev/adX.eli., so I don't think that's it. =C2=A0I've seen the problem wi= th > Samba, NFS, and dd. David, do you think you are willing to re-create the problem and do a PMC analysis on it? (If you need any guidance let me know, I will be happy to give it). Attilio --=20 Peace can only be achieved by understanding - A. Einstein
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?t2o3bbf2fe11004200529h45cf209et731a194794aae7e1>