Date: Tue, 13 Sep 2011 11:56:41 +0200 (CEST) From: Laszlo KAROLYI <szoftos@freemail.hu> To: freebsd-fs <freebsd-fs@freebsd.org> Subject: Re: ZFS-lighttpd2-sendfile, too high IO Message-ID: <freemail.20110913115641.67732.1@xmldata08.freemail.hu> In-Reply-To: <alpine.GSO.2.01.1109121428570.12948@freddy.simplesystems.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Hello,=0A=0Ait seems that turning off sendfile() in lighttpd2 completely so= lved my problem.=0A=0AThe inactive memory usage ceased, wired memory raised= for ARC cache, and I see an enormous fall of IO load under the same (or ev= en higher) network load. The disk IO load now scales to the network load.= =0A=0ASomething must be not right with the sendfile support in ZFS.=0A=0AHo= wever, my problem solved, and I'd suggest turning off sendfile in any progr= ams in the future, unless this bug gets fixed.=0A=0ALaszlo=0A=0ABob Friesen= hahn <bfriesen@simple.dallas.tx.us> =EDrta:=0A>On Mon, 12 Sep 2011, Laszlo = KAROLYI wrote:=0A>=0A>> But does this explain the 4-5mbyte/s reads when hav= ing a 15mbit/s network load?=0A>=0A>There are only two viable explanations:= =0A>=0A> o Insuffient caching due to insufficient resources=0A>=0A> o D= ata is not being cached at all=0A>=0A>Zfs reads whole 128K blocks (or whate= ver the filesystem blocksize is) =0A>at a time. It does not read partial b= locks from underlying storage. =0A>This makes it very expensive to perform = many small read accesses if =0A>the reads are not subsequently cached in th= e ARC.=0A>=0A>Bob=0A>-- =0A>Bob Friesenhahn=0A>bfriesen@simple.dallas.tx.us= , http://www.simplesystems.org/users/bfriesen/=0A>GraphicsMagick Maintainer= , http://www.GraphicsMagick.org/=0A
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?freemail.20110913115641.67732.1>