[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Solid State Hard Drives



On Fri, Oct 09, 2009 at 01:06:12AM +0200, Michelle Konzack wrote:
> Am 2009-10-08 11:06:39, schrieb Glenn Hocking:
> > Hi List
> > 
> > Had anyone had experience using Solid State Hard disks in mail servers?
> > I've heard that Solid State is very fast but not good for high read/write
> > functions such as mail servers as the drives wear out over time.
> 
> I can not believe, that a 128 GByte SSD is FASTER then a 147  GByte  SAS
> and the SAS cost me 230 Euro (my price as ISP), the SSD over 400...

It all depends on *which* SSD. 

Someone said buy SLC. That's good advice. But, even within SLCs there
are big differences in performance. A lot depends on how smart the controller
embedded in the drive is. Even more depends on what it is optimized for. Most
SSDs are optimized for maximum sequential read and write. That's a shame,
because there are very few real world workloads that can benefit from that.
Intel does it differently. Their drives are optimized for random read/write
performance.

This article is more than 6 months old but explains this well, with real
performance numbers. On the upside, the drives have come down in price a lot
since then:

  http://www.anandtech.com/printarticle.aspx?i=3531

I would say buy Intel X25M, or, if your budget permits, X25E. Note that the
X25M is MLC but with random write performance that is way, way better than a
lot of SLCs out there.

There are other vendors that have high-performance offerings, but I think
you'll find them way more expensive and/or not really released yet. See for
instance:

  http://www.computerworld.com/s/article/9137909/Start_up_releases_uber_fast_efficient_enterprise_class_SSDs?taxonomyId=19

An X25E will *blow your spinning platter SAS drive out of the water*. I
suspect that even an X25M might, for most workloads.

Thanks,
Ward.

-- 
Pong.be         -(   "Microsoft isn't evil, they just make really crappy   )-
Virtual hosting -(              operating systems." -- Linus               )-
http://pong.be  -(                                                         )-
GnuPG public key: http://pgp.mit.edu


Reply to: