From owner-freebsd-questions@freebsd.org Thu Jan 21 23:28:19 2016 Return-Path: Delivered-To: freebsd-questions@mailman.ysv.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:1900:2254:206a::19:1]) by mailman.ysv.freebsd.org (Postfix) with ESMTP id 20383A8B9AC for ; Thu, 21 Jan 2016 23:28:19 +0000 (UTC) (envelope-from mason@blisses.org) Received: from phlegethon.blisses.org (phlegethon.blisses.org [50.56.97.101]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (Client did not present a certificate) by mx1.freebsd.org (Postfix) with ESMTPS id 0AC8C1E7F for ; Thu, 21 Jan 2016 23:28:18 +0000 (UTC) (envelope-from mason@blisses.org) Received: from blisses.org (cocytus.blisses.org [23.25.209.73]) by phlegethon.blisses.org (Postfix) with ESMTPSA id 2DC9B1493BE; Thu, 21 Jan 2016 18:28:12 -0500 (EST) Date: Thu, 21 Jan 2016 18:28:10 -0500 From: Mason Loring Bliss To: dweimer Cc: freebsd-questions@freebsd.org Subject: Re: ZFS performance help sought Message-ID: <20160121232810.GJ4538@blisses.org> References: <20160121205139.GG4538@blisses.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.23 (2014-03-12) X-BeenThere: freebsd-questions@freebsd.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: User questions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 21 Jan 2016 23:28:19 -0000 On Thu, Jan 21, 2016 at 03:35:23PM -0600, dweimer wrote: > Try Starting here: > https://wiki.freebsd.org/ZFSTuningGuide#General_Tuning I'd looked at that previously, but as noted, I'd already tried capping the ARC, and txg.timeout is already five by default in FreeBSD. My reading about prefetch_disable suggests that it's meant for big transactions like what I'm doing, but I could be confused. All that said, I wasn't setting write_limit_override, so I'm trying that, and I'm cutting the txg.timeout back a couple seconds more. Am I confused about prefetch, though? They're talking about it as being problematic for small, random I/O operations, where I'm doing big, chunky I/O operations. Anyway, thanks for the ideas. I'll write back with results presently. -- Mason Loring Bliss (( If I have not seen as far as others, it is because mason@blisses.org )) giants were standing on my shoulders. - Hal Abelson