Message-ID: <4DBB0B76.8090805@2ndquadrant.com>
Date: Fri, 29 Apr 2011 15:03:18 -0400
From: Greg Smith <greg@2ndquadrant.com>
User-Agent: Thunderbird 2.0.0.24 (X11/20100317)
MIME-Version: 1.0
To: Ben Chobot <bench@silentmedia.com>
CC: Mark Steben <mark.steben@autorevenue.com>,
	pgsql-performance@postgresql.org
Subject: Re: FUSION-IO io cards
References: <BANLkTikdOoUZyvcNEXk67563tf7c78ZGUQ@mail.gmail.com>
	<69D96B03-F991-4825-A8FF-0FEEDCC17461@silentmedia.com>
In-Reply-To: <69D96B03-F991-4825-A8FF-0FEEDCC17461@silentmedia.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Ben Chobot wrote:
> Also, while I would say they seem reliable (they have a supercap and 
> succeeded every power-pull test we did) we just recently we've had 
> some issues which /appear/ to be fio driver-related that effectively 
> brought our server down. Fusion thinks its our kernel parameters, but 
> we're unconvinced, given the length of time we've run with the same 
> kernel settings. I'm not yet ready to say these cards are unreliable, 
> but I'm no longer willing to say they're problem-free, either.

Ben has written a nice summary of the broader experience of everyone 
I've talked to who has deployed Fusion IO.  The race car anology is a 
good one.  Fast, minimal concerns about data loss, but occasional quirky 
things that are frustrating to track down and eliminate.  Not much 
transparency in terms of what it's doing under the hood, which makes 
long-term reliability a concern too.  A particularly regular complaint 
is that there are situations where the card can requite a long 
consistency check time on system boot after a crash.  Nothing lost, but 
a long (many minutes) delay before the server is functioning again is 
possible.

The already mentioned TI RAMSAN at an ever higher price point is also a 
possibility.  Another more recent direct competitor to FusionIO's  
products comes from Virident:  http://www.virident.com/  They seem to be 
doing everything right to make a FusionIO competitor at the same basic 
price point.  They've already released good MySQL performance numbers, 
and they tell me that PostgreSQL ones are done but just not published 
yet; going through validation still.  The "Performance Relative to 
Capacity Used" graph at 
http://www.ssdperformanceblog.com/2010/12/write-performance-on-virident-tachion-card/  
is one anyone deploying on FusionIO should also be aware of.

-- 
Greg Smith   2ndQuadrant US    greg@2ndQuadrant.com   Baltimore, MD
PostgreSQL Training, Services, and 24x7 Support  www.2ndQuadrant.us
"PostgreSQL 9.0 High Performance": http://www.2ndQuadrant.com/books