Re: 3ware trivia overload - Mailing list pgsql-performance

I'll throw in my 2 cents worth:

1) Performance using RAID 1 for reads sucks. You would expect throughput to
double in this configuration, but it doesn't. That said, performance for
RAID 1 is not noticeably worse than Linux MD. My testing showed the 3Ware
controller to be about 20% faster than Linux MD for RAID 1.

2) The management tools work well and can keep you informed of problems with
the array. They're also web based, so they are easy to use from other
computers. Unlike some systems like IBM's which is Java based, or Adaptec's
which just plain don't work on Linux (at least I could never get it to
work).

3) They are very reliable. I've never had a problem I could attribute to the
controller.

4) I recently did some testing with a single SSD drive and a 3Ware
9550SXU-4LP controller. I got something like 3,000 IOPs per second for
random writes. I think the problem is not so much that the 3Ware controller
sucks as that Consumer Grade SATA drives suck.

5) A common problem is that if you don't have the BBU, the write cache
defaults to disabled. This can just kill performance. You have to either
purchase a BBU or live dangerously and enable the write cache without the
BBU.

6) A lot of the performance numbers bandied about in those tests are not
valid. For example, one test was done using a 385 MB file, which is almost
certainly smaller than RAM on the system. One guy talks about getting 6000
random reads per second from a standard consumer grade SATA drive. That's a
seek time of .0166 ms. Not realistic. Clearly the Linux Page cache was
involved and skewing numbers.

So, my opinion is that if you want really reliable RAID performance using
consumer grade drives, the 3Ware controller is the way to go. If reliability
is not so important, use Linux MD. If you want high performance pay the very
high $/GB for SCSI or SSD.

George Sexton
MH Software, Inc.
303 438-9585
www.mhsoftware.com


> -----Original Message-----
> From: pgsql-performance-owner@postgresql.org [mailto:pgsql-performance-
> owner@postgresql.org] On Behalf Of Greg Smith
> Sent: Saturday, September 18, 2010 1:51 AM
> To: pgsql-performance@postgresql.org
> Subject: [PERFORM] 3ware trivia overload
>
> Since I've never stumbled on this until now, and I spend all day
> looking
> at this sort of thing, I'm guessing many of you haven't seen the
> following collection of 3ware controller trivia either:
> http://makarevitch.org/rant/3ware/
>
> All moving toward obsolete with 3ware being a part of LSI now I
> suspect,
> but a great historical piece.  It puts some numbers on the general
> anecdotal observations that the older 9500 series of cards from 3ware
> were terrible in several common situations, and that the 9600 and later
> cards aren't so bad.  I've never seen so much of the trivia related to
> using these particular cards assembled into one well referenced place
> before.  Probably interesting to those interested in general Linux disk
> performance issues even if you don't care about this particular brand
> of
> card.
>
> --
> Greg Smith, 2ndQuadrant US greg@2ndQuadrant.com Baltimore, MD
> PostgreSQL Training, Services and Support  www.2ndQuadrant.us
> Author, "PostgreSQL 9.0 High Performance"    Pre-ordering at:
> https://www.packtpub.com/postgresql-9-0-high-performance/book
>
>
> --
> Sent via pgsql-performance mailing list (pgsql-
> performance@postgresql.org)
> To make changes to your subscription:
> http://www.postgresql.org/mailpref/pgsql-performance