Re: Repeatable kernel panic on -CURRENT using ZFS over SATA

From: 韓家標 Bill Hacker <askbill_at_conducive.net>
Date: Tue, 02 Oct 2007 16:15:26 -0400
Steven Schlansker wrote:
> Pawel Jakub Dawidek wrote:
>> On Tue, Oct 02, 2007 at 01:17:00AM -0700, Steven Schlansker wrote:
>>> Hello everyone,
>>> I recently set up a 6 drive SATA raidz2.  Whenever I try to use the 
>>> array, the dmesg fills up with warnings that WRITE_DMA must be 
>>> retried (repeatedly)
>>>
>>> As soon as I remove the load, everything runs fine.
>>>
>>> Dmesg with errors here:
>>> http://soda.csua.berkeley.edu/~steven/dmesg.txt
>>>

*trimmed*

 >>> The eventual end result:
 >>> http://soda.csua.berkeley.edu/~steven/Image053.jpg
 >>>
 >>>
 >>> The only references I can find to similar problems were either not
 >>> resolved, or seemed to be related to a chipset which I am not using.
 >>>
 >>> Is this a known issue?  How can I make this machine stable?

*trimmed*

Short answer - you are overstressing your very marginal hardware.

Long answer - here's how and why:

Your rig was state of the art 7+ years ago:

http://www.supermicro.com/newsroom/pressreleases/2000/press101800.cfm

Though Dell bulds to a prce point, and may have cut a few corners..

First, your power supply may not be remaining stable under startup load:

- the 6 SeaGraaates need over 200 W to spin up, about 80W to run, and the ATA 
controllers may not be staggered-start capable.

- the other two SeaGraates and lone Fujitsu on SCSI need another 55 W or so to 
start up and 30+W for running load, but *are* programmable for delayed/staggered 
spin-up. Do so if you have not already...

Then try this:

Interrupt the boot. Give the rig a few minutes to spin-up, settle down, cool 
down, stabilize. Then continue the boot.


Other factors:

- 7+ year old el-cheapo (Adaptec) SCSI controller was marginal when brand-new.

- The el-cheapo Silicon Image and onboard Intel SATA OTOH are a far tougher 
challenge. These rely very heavily on the system RAM, CPU, and bus to do 
everything 'arrayish'.

And there is the challenge:

There are very few true 'hardware' SATA controllers made.  Most decent ones are 
combo SATA+iSCSI, cost the very Earth, (US$ 700 UP) and are rare enouhg in new 
channels, let alone surplus.

Bottom line is that the rig you are using hasn't got the resources for the 
massive storage you are trying to mount.

When such storage was to be found only on 'Big iron' - even the older, slower 
Power, Alpha, MIPS or SPARC IBM/DEC/Sun 'not-so-big' iron, very powerful 
peripheral controllers carried much of the load.

A dual Pentium III 733 MHz with only 512 MB of (quite slow) RAM, marginal SCSI 
controller, and bottom-end resource-hungry Silicon Image/Intel SATA is not even 
on the same *planet* as those old beasts, and a big fat enclosure with a Dell 
logo changeth that not.


FreeBSD is superbly efficient. But not quite magical.

If you want the rig to be stable, you will either have to uprade it 
substantially, ELSE downshift the I/O to do everything at a more leisurely pace.

Not what you wanted to hear, I'm sure...

Bill 'hardware' Hacker
Received on Tue Oct 02 2007 - 18:26:25 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:18 UTC