From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ryan Subject: (unknown) Date: Sun, 18 Apr 2004 18:24:45 -0500 Sender: linux-ide-owner@vger.kernel.org Message-ID: <1082330685.40830e3da9c73@webmail.LaTech.edu> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7BIT Return-path: Received: from selene.LaTech.edu ([138.47.18.25]:44487 "EHLO LaTech.edu") by vger.kernel.org with ESMTP id S264211AbUDRXYr (ORCPT ); Sun, 18 Apr 2004 19:24:47 -0400 Received: from localhost (localhost [127.0.0.1]) by LaTech.edu (Postfix) with ESMTP id 5114330A060 for ; Sun, 18 Apr 2004 18:24:46 -0500 (CDT) Received: from LaTech.edu ([127.0.0.1]) by localhost (selene.latech.edu [127.0.0.1]) (amavisd-new, port 10024) with LMTP id 83671-01-17 for ; Sun, 18 Apr 2004 18:24:45 -0500 (CDT) Received: from localhost (helius.LaTech.edu [138.47.18.18]) by LaTech.edu (Postfix) with ESMTP id B60B1309FB5 for ; Sun, 18 Apr 2004 18:24:45 -0500 (CDT) List-Id: linux-ide@vger.kernel.org To: linux-ide@vger.kernel.org > > > >Promise's S150 SX4 SATA controller supports hotswap and automatic array > >rebuilds. > > > Really? How long has it done that? The last libata status report I saw > from Jeff Garzik said hotswap was still unsupported. Has that changed? > Do you know if any other interface card's driver supports it yet? It > supports that standard SCSI hot-plug utilities or what? > I said the CARD supports it, not the drivers. Perhaps I was a bit unclear on that. Currently the SX4 is supported through the sata_promise driver. It has NO support for any controller-level RAID or drive hotswapping. Jeff Garzik, I think, is working on the SX4 driver. As far as I know, with the SX4, RAID rebuilding and hotswapping within an array should be handled at the hardware level by the card (at least for RAID5 arrays), but I really don't know that much about how it all fits together. I could very well be wrong. > >The problem with it is that at the moment there is no driver support > >for the SX4's RAID functionality. So in the meantime I'm using Linux's > >software RAID5 on the card. It's really quite slow (25mb/s) but that's > mostly > >due to the current driver support, from what I've been told. > > > > > I run RAID1, I trade space for performance and simplicity. I run > latency sensitive stuff, so RAID1's read performance is what I need. It > lowered average latency 40% and increased seeks/sec by 80% as compared > to a single disk setup. Plus it'll do parallel sequential reads at > 100MB/s, if I had sequential access patterns. It doesn't appear to have > any noticeable affect on CPU usage either. >>From what Jeff said, drive access is slowed considerably on the SX4 due to how the current driver accesses the disk. And I quote: "SX4 is vastly different. Each request must bounce through a DIMM, which hurts performance. Further, only one DIMM copy to/from system memory can be occuring at any one time, while you can be executing up to 4 ATA commands (one for each SATA port) in parallel." --Jeff Garzik So no matter what RAID array I use, I'm greatly limited by the current driver design. Like I said, 25mb/s on a single drive, and I'm geting about 29mb/s for the whole array. I'd expect around an 80mb/s read on the array with the "ideal" drivers. And some more from Jeff: "SX4 is really designed for RAID acceleration. One may to improve performance (which I plan to implement) is using the DIMM as a write-through cache." Drive performance can really be split up into a couple of different areas: Disk-level, controller-level, and driver-level. Some disks are inherently faster than others. Some controllers are better than others at certain things (the SX4 is SUPPOSED to be a RAID card, not a simple controller card) and then, drivers can either suck or not. I guess you could say RAID is either controller-level, driver-level, or both. Personally, I have to commend the guys working on the Promise SATA driver(s), I think they've done a great job getting this all together. I'm looking forward to seeing an SX4 driver that supports all of the wonderful things the card can do. > > -ryan > -- huh? ------------------------------------------------- This mail sent through IMP: http://horde.org/imp/