[zfs-discuss] Re: 6x drive raidz1 Ubuntu 12.04 performance problems

Gordan Bobic gordan.bobic at gmail.com
Sun Jun 3 05:17:29 EDT 2012


On 06/02/2012 06:06 PM, DC wrote:
> Wow - good information here.
>
> I don't think it's the power-supply.  This particular setup has been
> running without any problems for over a year, and has been nice and
> zippy.  I did switch the cpu DOWN from a quad-core to a dual-core
> about six months ago, and noticed a performance hit, but nothing truly
> major.  It's only in the last couple of months that performance has
> dropped precipitously, something I attributed to the failing boot
> drive.  I swapped that out, upgrading in the process from 11.10 to
> 12.04 (and up to kernel 3.2 btw).

PSUs can fail gradually, with voltages drifting further and further out 
until things either stop working (voltage too low) or blow something up 
(voltage too high).

I thought it was also common knowledge that most PSUs are absolutely 
crap, and the labels on them in terms of power ratings are largely 
meaningless. Most cannot stably provide the power output they are 
allegedly rated to (you shouldn't aim to use more than 2/3 of the power 
they claim to be good for) and even then the power quality can be very 
low (noisy) which can cause all sorts of stability issues. You need an 
oscilloscope and a good load generator to benchmark a PSU.

I tend to stil with Coolermaster and PC Power and Cooling PSUs, based on 
past experience and test results, but things to bear in mind are:

1) They are a lot more expensive that others, including highly branded ones.
2) The quality can very massively between different models of the same 
brand.

It is also very difficult to find PSU benchmarks/reviews online that are 
particularly meaningful because most don't measure the noise on the 
output lines and how it varies with the load.

> Smartmontools doesn't really show any problems - no reallocated
> sectors etc.  And you can see from the munin graphs that the counts
> haven't changed.  Interestingly, sda and sdc have both had some
> Seek_Error_Rate hits on them ...

They reallocated sector counts won't change on most WD and Samsung 
disks. They cynic in me thinks that this was to reduce claims within the 
warranty period. In most big data (videos, music) corruption isn't 
always immediately obvious, and it's easy to ignore a small glitch. And 
if its an OS file that gets corrupted it's too easy to blame Windows or 
crap software.

> I also realized that I never configured /etc/smartd.conf specifically,
> so it's only using the usual DEVICESCAN method and not running any
> tests.  Does anyone have a sample smartd.conf config that they like to
> use with ZFS ?

I normally have a daily short self-test and a weekly long self-test. 
Unfortunately with my LSI 8-port SAS card (driving SATA disks), running 
SMART self-tests causes them to drop out. This appears to be a known, 
long standing driver/firmware bug that nobody appears to be doing 
anything about. The only option I have is to not run the self-tests on 
the disks attached to that controller and instead rely on ZFS scrubs 
(also weekly) and the disks' Write-Read-Verify feature to keep my data safe.

Gordan



More information about the zfs-discuss mailing list