Date: Tue, 16 Dec 2003 12:41:52 -0500 From: Beric Farmer <bfarmer@xe.com> To: Dan Melomedman <dan@devonit.com> Cc: freebsd-hardware@freebsd.org Subject: Re: SATA RAID controllers Message-ID: <531664931.1071578512@[192.168.1.100]> In-Reply-To: <20031208190253.GC19941@mail.devonit.com> References: <20031208190253.GC19941@mail.devonit.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Hi Dan. I've had experience with the HighPoint RocketRAID 1520 (SATA) and the Promise TX2 (ATA133). First, in terms of reliability and function, I've not had any problems with either to date. However, I find the transfer rates of both of them to be less than I would expect. I'm not sure what version of FreeBSD you're planning to use, but my understanding that SATA support isn't great in 4.x. Under 4.x the ata driver downgrades the SATA drives on the HighPoint to UDMA33 (33 MB/s) and I see read throughput of approximately 15 MB/s and write throughput of about 25 MB/s (whether I use FreeBSD's ata driver or the one that HighPoint provides). Under 5.1, the drives are run at UDMA133 (133 MB/s). (I think I read somewhere that the HighPoint controller is essentially an ATA133 controller with a bridge that translates the SATA150 to ATA133, so it doesn't actually run at 150 MB/s.) However, my throughput is actually worse than under 4.x. I see 15MB/s both reading and writing. This is way below what it should be. I've been in contact with the person who maintains FreeBSD's ata driver, and it appears that although his driver follows the specs that are available to him, the controller/drives don't behave as advertised. If you use atacontrol to drop the drives to UDMA100, the throughput improves significantly. In my case, it jumped to 53 MB/s reading and 43 MB/s writing. This is still somewhat slower than what I see with a single ATA133 drive on a plain ATA controller, but it's in the same ballpark. I realize the Promise TX2 isn't a SATA controller (its an ATA133 controller), but my experiences may still be helpful. With this controller, under 4.9-RELEASE, I see 40 MB/s reading and 20 MB/s writing, which is far lower than a single ATA133 drive. I'm not sure why it is low, but it appears that there is a bottleneck of a total of 40MB/s between the controller and its drives. This would explain why the write throughput is exactly half the read throughput (each piece of data can be read from a single drive, but must be written simultaneously to both drives). I'd be interested to hear from other people regarding ATA RAID 1 implementations in general. I've read that RAID 1 gives the ability to achieve double the read transfer rates of a single drive because the reads can be distributed across the mirrored pair. However, the data transfer rates on the controllers I've mentioned above suggest they don't take advantage of this (and tests with iostat during a large transfer suggest that all the data is indeed being read from a single drive). Has anyone had experience with an ATA (or SATA) RAID 1 controller that seems to take advantage of this possibility of RAID 1? Perhaps the manufacturers assume that anyone concerned with throughput will be using a different RAID level, so don't bother to go to the expense... Hope this info is helpful. Regards, Beric --On December 08, 2003 14:02 -0500 Dan Melomedman <dan@devonit.com> wrote: > > I am looking for a cheap well-supported SATA RAID card without > performance glitches in RAID 1 configuration. > > Here are possible candidates: > > Promise FastTrak S150 TX2+, TX4 > LSI Logic MegaRAID SATA 150-2 Serial ATA Raid > Highpoint RocketRAID 1520 or 1540 or 1640 > Adaptec Serial ATA Raid 1210SA (I've read somewhere that Adaptec's > card suffer from poor performance, is this still true for this > model?) Acard cards? >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?531664931.1071578512>