[ale] Build-yer-own NAS server

[Greg Freemyer]

On 6/8/06, J. D. Pearson <jpearson at turbocorp.com> wrote:
>
> >> I have been avoiding SATA with Linux due to the poor EH capability.
> >> WIth the new EH, I believe Linux/Sata will finally be ready for
> >> production use.  OTOH, it may take a few more months to shake out the
> >> new EH.  So I will wait to see how stable it appears to be once it
> >> comes out.
> >
> >
>
> I agree. We ran into problems recently with some SATAII drives that
> seemed to be
> related to Native Command Queuing feature.  When the support  is there
> for the newer drives with
> NCQ it seems they might be the obvious choice for a home server with
> newer hardware. What
> do you think?
>
> Best regards,
>
> J. D.

Background:
NCQ allows multiple outstanding write requests to live on the drive,
but still allows write-barriers to be supported as needed for
journaled FS.  I don't understand the details, but without NCQ
journelled FSes generally recommend turning off write-cache in the
drive so you can only have one write request on the drive at a time.

Back on topic:
There has been some discussion on the sata kernel list about NCQ.
They have it working in the new big rewrite (NCQ may not get in until
2.6.19), but benchmarking is showing it causes a small performance
drop for many (most?) workloads!!!.

Per one of the developers, NCQ makes a big difference if the OS does a
poor job of using a elevator i/o scheduler like a "certain OS", but
Linux 2.6 has a very good elevator i/o scheduler so you actually see a
drop in performance when NCQ is enabled for general workloads.

So, I don't think it is a feature I will use.

Greg
-- 
Greg Freemyer
The Norcross Group
Forensics for the 21st Century