From owner-freebsd-questions@freebsd.org Tue Aug 29 22:21:11 2017 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 5D11EDE707A for ; Tue, 29 Aug 2017 22:21:11 +0000 (UTC) (envelope-from cyberleo@cyberleo.net) Received: from mail.cyberleo.net (paka.cyberleo.net [216.226.128.180]) by mx1.freebsd.org (Postfix) with ESMTP id 4055C73274 for ; Tue, 29 Aug 2017 22:21:10 +0000 (UTC) (envelope-from cyberleo@cyberleo.net) Received: from [172.16.44.4] (vitani.den.cyberleo.net [216.80.73.130]) by mail.cyberleo.net (Postfix) with ESMTPSA id 66A2D2B402; Tue, 29 Aug 2017 18:11:27 -0400 (EDT) Subject: Re: NFS home directory performance tuning for Linux client To: Kaya Saman , freebsd-questions@freebsd.org References: <30d13a2b-0813-9686-3841-b24051fa3e0e@gmail.com> From: CyberLeo Kitsana Message-ID: <81615b0c-a94c-1fca-de5c-cc8cc868b05c@cyberleo.net> Date: Tue, 29 Aug 2017 17:11:26 -0500 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Thunderbird/52.2.0 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset=utf-8 Content-Language: en-GB Content-Transfer-Encoding: 8bit X-BeenThere: freebsd-questions@freebsd.org X-Mailman-Version: 2.1.23 Precedence: list List-Id: User questions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 29 Aug 2017 22:21:11 -0000 On 08/21/2017 01:47 PM, Kaya Saman wrote: > So, currently I've tried doing something a little different which worked > out well. > The bottleneck was definitely caused by NFS and I think it was the write > behaviour for small files, with the limited options available though I > have no idea what could be causing the issues or how to get round them?? Try setting sync=disabled on the zfs datasets backing your NFS shares. If this has a noticeable impact on performance, you may instead want to invest in a SLOG device. This can be a mirrored pair of high-iops/low-latency SSDs, or a battery-backed RAM device. Ordinarily, the ZIL is on the pool's main storage. If you have a bunch of high-latency spinning disks, this can slow sync writes way down as they then have to wait for the disks to finish. An external ZIL on a fast SLOG device will boost sync writes closer to the throughput of the SLOG than the pool. If I remember correctly, all NFS writes are sync writes, and will be impacted by the latency of the ZIL. Setting sync to disabled on a dataset will eliminate the guarantees of sync writes (data loss possible in the case of crash or power loss) but may speed up NFS through this compromise. I would not recommend leaving sync disabled, though. https://www.ixsystems.com/blog/o-slog-not-slog-best-configure-zfs-intent-log/ -- Fuzzy love, -CyberLeo Technical Administrator CyberLeo.Net Webhosting http://www.CyberLeo.Net Element9 Communications http://www.Element9.net Furry Peace! - http://www.fur.com/peace/