Received: from maia.hub.org (maia-2.hub.org [200.46.204.251]) by mail.postgresql.org (Postfix) with ESMTP id CA3961337B67 for ; Fri, 29 Apr 2011 13:34:46 -0300 (ADT) Received: from mail.postgresql.org ([200.46.204.86]) by maia.hub.org (mx1.hub.org [200.46.204.251]) (amavisd-maia, port 10024) with ESMTP id 32684-02 for ; Fri, 29 Apr 2011 16:34:29 +0000 (UTC) X-Greylist: delayed 00:30:08.761022 by SQLgrey-1.7.6 Received: from mail1.iathh.de (mail1.iathh.de [217.89.104.238]) by mail.postgresql.org (Postfix) with ESMTP id 8804B1337999 for ; Fri, 29 Apr 2011 13:34:28 -0300 (ADT) Received: from [10.22.0.10] ([10.22.0.10]) (authenticated user joachim.worringen@iathh.de) by mail1.iathh.de (Kerio Connect 7.1.4); Fri, 29 Apr 2011 18:04:16 +0200 Message-ID: <4DBAE181.2000204@iathh.de> Date: Fri, 29 Apr 2011 18:04:17 +0200 From: Joachim Worringen Organization: IAT GmbH User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9.2.9) Gecko/20100922 Thunderbird/3.1.4 MIME-Version: 1.0 To: pgsql-performance@postgresql.org Subject: Re: FUSION-IO io cards References: <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 X-Virus-Scanned: Maia Mailguard 1.0.1 X-Spam-Status: No, hits=-1.91 tagged_above=-10 required=5 tests=BAYES_00=-1.9, T_RP_MATCHES_RCVD=-0.01 X-Spam-Level: X-Archive-Number: 201104/414 X-Sequence-Number: 43476 On 04/29/2011 04:54 PM, Ben Chobot wrote: > We have a bunch of their cards, purchased when we were still on 8.1 and > were having difficulty with vacuums. (Duh.) They helped out a bunch for > that. They're fast, no question about it. Each FusionIO device (they > have cards with multiple devices) can do ~100k iops. So that's nifty. > > On the downside, they're also somewhat exotic, in that they need special > kernel drivers, so they're not as easy as just buying a bunch of drives. > More negatively, they're $$$. And even more negatively, their drivers > are inefficient - expect to dedicate a CPU core to doing whatever they > need done. I would recommend to have a look a Texas Memory Systems for a comparison. FusionIO does a lot of work in software, as Ben noted correctly, while TMS (their stuff is called RAMSAN) is a more all-in-hardware device. Haven't used TMS myself, but talked to people who do know and their experience with both products is that TMS is problem-free and has a more deterministic performance. And I have in fact benchmarked FusionIO and observed non-deterministic performance, which means performance goes down siginificantly on occasion - probably because some software-based house-keeping needs to be done. -- Joachim Worringen Senior Performance Architect International Algorithmic Trading GmbH