Re: [vserver] SSD/HD hybrids

From: Eugen Leitl <eugen_at_leitl.org>
Date: Fri 02 Apr 2010 - 19:38:02 BST
Message-ID: <20100402183802.GU1964@leitl.org>

On Fri, Apr 02, 2010 at 04:19:26PM +0900, Christian Balzer wrote:
> On Thu, 1 Apr 2010 11:32:16 +0200 Eugen Leitl wrote:
> >
> > So I'm thinking about how to use such a hybrid SSD/HD setup manually.
> > The obvious solutions would be to use the SSD for swap and for /tmp,
> > and HD for everything else.
> >
> I would be _very_ worried about using a single SSD, especially a MLC one

Of course I would like to use an SLC or two of them. Way out of budget
for now, though.

> for something that you clearly think will have LOTS of write accesses to

Not necessarily (I'll also use noatime in /etc/fstab options). In fact I'm
considering using 2x X-25M Intel SSDs in RAID 0. The rationale is that
this should double the write endurance at only slightly increased
(intrinsic) failure rate (I have zero failure data on SSDs so far, the population of
those 15-20 I have in production haven't failed yet) while pushing
the performance into utterly ridiculous and unbalanced (in regards
to CPU, dual-core Atom is pretty weak sauce) territory.

The vserver guests are all rsync-mirrored nightly (or more frequently,
should this become necessary) to a raidz2 or raidz3 box, and I'd rather
have RAIS (redundant array of independent servers) than multiple, expensive
redundancies within the server itself.

> it. But then again, looking at your HW description you don't seem to be
> overly worried about running with just one fallible spinning thing either.

I can fit in 2x 2.5". I'm thinking of a pair of 500 GByte Segate Constellation
or the just-announced 600 GByte VelociRaptor. Out of budget, for time being.
I've seen 1U short rackmounts from SuperMicro which take 4x 2.5" SATA/SAS
frontal hot-plug however.
 
> For reference, I just recently did build our first DRBD cluster with the
> OS (/var and most importantly DRBD metadata partitions) on Intel SLC SSDs.
> Yes, plural, I'm running 2 of them in a RAID1 for better sleep AND have
> spare partitions (Linux MD) standing by on a "real" disk. The performance
> gain was still very much worth it in our case and the cost versus my
> aforementioned better sleep at night just as well.

I'm trying to not have multiple points of failure. Once the cash flow
is there I can buy unobtainium components, not before.
 
> And yes, BTRFS would be nice to have but I'm not holding my breath on that
> one either.

We agree.

-- 
Eugen* Leitl leitl http://leitl.org
______________________________________________________________
ICBM: 48.07100, 11.36820 http://www.ativel.com http://postbiota.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE
Received on Fri Apr 2 19:38:17 2010
[Next/Previous Months] [Main vserver Project Homepage] [Howto Subscribe/Unsubscribe] [Paul Sladen's vserver stuff]
Generated on Fri 02 Apr 2010 - 19:38:19 BST by hypermail 2.1.8