From mboxrd@z Thu Jan 1 00:00:00 1970 From: Konstantin Olchanski Subject: Re: SATA performace Date: Mon, 2 Jan 2006 21:04:15 -0800 Message-ID: <20060103050414.GF28438@sam.triumf.ca> References: <17336.42843.779496.644030@smtp.charter.net> <000d01c61017$ea630d10$3224050a@avilespaxp> <20060103044759.GA8876@lug.udel.edu> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: <20060103044759.GA8876@lug.udel.edu> Sender: linux-raid-owner@vger.kernel.org To: Ross Vandegrift Cc: Paul Aviles , linux-raid@vger.kernel.org List-Id: linux-raid.ids On Mon, Jan 02, 2006 at 11:47:59PM -0500, Ross Vandegrift wrote: > On Mon, Jan 02, 2006 at 10:43:47PM -0500, Paul Aviles wrote: > > Are SATA drives similar in performance than IDE drives? I have tested > > Barracudas 7200.0 (500Gb) and WD too on the same type of servers (more than > > 1 unit) and what I am getting is painfully slow in terms of read/writes. In modern systems, disk performance is dominated by the physical disk characteristics: rotation speed and seek times (ignoring speedups from Linux-level, controller-level and disk-level i/o reordering and caching). So you should (and we do) see almost identical performance between similar PATA and SATA disks. For bulk data streaming, you should see 30-60 Mbytes/sec for a single disk. The only performance-degrading problems I have seen are a) PATA disks running in non-DMA modes (bulk data streaming rate = 3 Mbytes/sec) and b) barely-readable sectors on many new high-density disks (disk-level read retries, takes seconds to read one sector, kills performance). Problem (b) is quite evil and hard to diagnose: it seems to be temperature dependant, it is not reported by SMART, and it is not reported by Linux (unless it is so bad that you get a read timeouts). For RAID sets, it causes erratic performance. -- Konstantin Olchanski Data Acquisition Systems: The Bytes Must Flow! Email: olchansk-at-triumf-dot-ca Snail mail: 4004 Wesbrook Mall, TRIUMF, Vancouver, B.C., V6T 2A3, Canada