Re: The read and write timeouts story, next stage...

[Teemu Rinta-aho <teemu%rinta-aho.org@localhost>]

I did some more testing. I installed FreeBSD 8.0-RELEASE on the original
two Seagate drives, using gmirror to have a similar RAID-1 setup as
I had with NetBSD and RAIDframe.

I mirrored the other drive and run through "make -j4 universe" on my 
Core 2 server (Intel ICH8) without problems. With NetBSD I got problems
with RAIDframe parity checking. Then I moved the drives to the original 
Mini-ITX server (Intel ICH7) and ran "make -j8 universe" which lasted 
21+ hours and in parallel I installed and ran bonnie++ two times with 10 
loops each run.

I changed the geom algorithm from "round-robin" to "split" in the
middle of the 21 hour make session. I saw a lot of 90+ % loads on
both drives with 'gstatus'. Top showed loads of 7.x all the time.

I did not get any errors from any hardware component. That is the
exact same hardware setup that I can reproduce problems with NetBSD
when I run "build.sh release".

I have not yet studied the FreeBSD source code but it seems that
the software makes a difference here. I don't know if FreeBSD has
some cludges to bypass some hardware bugs. However, it seems to work
so far from the system admin point of view..

BR,
Teemu

On 20.12.2009 14:19, Teemu Rinta-aho wrote:
> Hi all,
>
> I have been reporting here my battle with the read and write timeouts
> from SATA drives. Now I managed to try the drives on a 3rd PC, where
> completely everything is different from the previous 2 PCs with which
> I have had the problem. And the problem is still there! (The third PC
> is about one month old, with Asus P7P55D motherboard, Core i7 860 CPU,
> 750W PSU etc.)
>
> It seems that I can reproduce write timeouts with the Intel Mini-ITX
> by building the NetBSD source. On Core 2 I get read timeouts when
> I let the raidframe do a parity check on the RAID-1 set. On Core i7
> I got a write timeout when I did a rebuild of the other drive in the
> RAID-1 set.
>
> Now I am bored of finding new pieces of hardware to try to make the
> problem go away and was wondering if anyone has a similar setup that
> I am having and could try if he or she can reproduce the problem.
>
> My setup is: two 1TB SATA2 drives configured to RAID-1 using
> raidframe. It doesn't seem to make a difference whether the driver
> is piixide or ahcisata, I have tried to change BIOS settings etc.
> I have all the NetBSD partitions on that RAID device, including
> root. Then by either building a NetBSD release or causing the
> raidframe to do parity check (by e.g. pressing the reset button)
> I can reproduce the problem.
>
> The only common nominators in my tests are: the power coming from
> the wall, the UPS, and the NetBSD code.
>
> I will try -current with the Core i7 machine next.
>
> BR,
> Teemu