Date: Tue, 04 Aug 2015 09:28:17 +0200 From: Alexander Leidinger <Alexander@leidinger.net> To: Matt Churchyard <matt.churchyard@userve.net> Cc: Quartz <quartz@sneakertech.com>, FreeBSD FS <freebsd-fs@freebsd.org> Subject: Re: ZFS: Disabling ARC? Message-ID: <20150804092817.Horde.aha5Ph2llansySR3xJXVMN2@webmail.leidinger.net> In-Reply-To: <45e74b89dd754991a366a46aa5101822@SERVER.ad.usd-group.com> References: <55BC14B7.9010009@sneakertech.com> <20150801133635.00002ecc@Leidinger.net> <45e74b89dd754991a366a46aa5101822@SERVER.ad.usd-group.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Quoting Matt Churchyard <matt.churchyard@userve.net> (from Mon, 3 Aug 2015 15:00:07 +0000): > I don't know if it's changed, but even changing ARC to metadata only > used to cause massive performance issues when reading large files in > small chunks. Reading a 128k ZFS record in 4k chunks would cause ZFS > to read the same 128k record from disk 32 times. > There's a forum thread about it here - > https://forums.freebsd.org/threads/zfs-primarycache-all-versus-metadata.45555/ When using "metadata" as the setting, you tell that you are OK to trade in some (data access, not metadata-access) performance for memory. Determining if the resulting performance is acceptable is up to you. If you blindly set the value without validating that it is OK for you, you shall not complain (= "measure, change, measure" and not "measure, change, hope"). I agree that not for all changes the resulting effects are 100% obvious and explicitely stating the edge-cases is appreciated. Feel free to add some words about it to https://wiki.freebsd.org/ZFSTuningGuide Bye, Alexander. -- http://www.Leidinger.net Alexander@Leidinger.net: PGP 0xC773696B3BAC17DC http://www.FreeBSD.org netchild@FreeBSD.org : PGP 0xC773696B3BAC17DC
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20150804092817.Horde.aha5Ph2llansySR3xJXVMN2>