Skip to contentSkip to conversation

tuning for large files in xfs

DISCUSSION fitzboy 1 participant 1 message May 23, 2006

lore
Share & tools
lore

fitzboy

author
#0 lore raw

Return-Path: <[email protected]>
X-Original-To: [email protected]
Delivered-To: [email protected]
Received: from horus.isnic.is (horus.isnic.is [193.4.58.12])

	(using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits))
	(No client certificate requested)
	by attila.bofh.it (Postfix) with ESMTP id 057865F87F
	for <[email protected]>; Tue, 23 May 2006 02:59:46 +0200 
(CEST)
Received: from proxy.google.com (proxy.google.com [66.102.0.4])
	by horus.isnic.is (8.12.9p2/8.12.9/isnic) with ESMTP id k4N0xduC097233
	for <[email protected]>; Tue, 23 May 2006 00:59:41 GMT
	(envelope-from [email protected])
Received: from  G018037
	by proxy.google.com with ESMTP id k4N0xXAA020036
	for <[email protected]>; Mon, 22 May 2006 17:59:33 -0700
Received: (from news@localhost)
	by Google Production with  id k4N0xXvP022170
	for [email protected]; Mon, 22 May 2006 17:59:33 -0700

To: [email protected]
Path: j73g2000cwa.googlegroups.com!not-for-mail
From: [email protected]
Newsgroups: linux.kernel
Subject: Re: tuning for large files in xfs
Date: 22 May 2006 17:59:27 -0700
Organization: http://groups.google.com
Lines: 12
Message-ID: <[email protected]>
References: <[email protected]> <[email protected]>
NNTP-Posting-Host: 209.209.36.196
MIME-Version: 1.0
Content-Type: text/plain; charset="iso-8859-1"
X-Trace: posting.google.com 1148345973 22160 127.0.0.1 (23 May 2006 00:59:33 GMT)
X-Complaints-To: [email protected]
NNTP-Posting-Date: Tue, 23 May 2006 00:59:33 +0000 (UTC)
In-Reply-To: <[email protected]>
User-Agent: G2/0.2
X-HTTP-Useragent: Mozilla/5.0 (Macintosh; U; PPC Mac OS X; en) AppleWebKit/418 (KHTML, like Gecko) Safari/417.9.2,gzip(gfe),gzip(gfe)
Complaints-To: [email protected]
Injection-Info: j73g2000cwa.googlegroups.com; posting-host=209.209.36.196;
posting-account=4elvtg0AAAAA6DoPt8oOAEoKYVdVm_2h

the sweet size for me would be a 32k block size on both the RAID and the XFS partition for me (that is the best number for my application). However, on the lower level RAID there is a very nominal performance difference between 32k and 64k stripe size (like 4%), so I just stick with the defauly 64k. And for the XFS partition, since I am on an intel machine with 2.6.8, I can only go up to a 2k blocksize...

But to answer your question, the RAID is on a 64k stripe size, and I just changed my test app to do 2k reads, and still I get the same performance (with only marginal improvement), so alignment can't be the