Date: Wed, 04 Jul 2012 23:59:35 +0400 From: "Alexander V. Chernikov" <melifaro@FreeBSD.org> To: lev@FreeBSD.org Cc: Doug Barton <dougb@FreeBSD.org>, net@freebsd.org Subject: Re: FreeBSD 10G forwarding performance @Intel Message-ID: <4FF4A0A7.9050601@FreeBSD.org> In-Reply-To: <17110267579.20120704233705@serebryakov.spb.ru> References: <4FF319A2.6070905@FreeBSD.org> <4FF361CA.4000506@FreeBSD.org> <20120703214419.GC92445@onelab2.iet.unipi.it> <4FF36438.2030902@FreeBSD.org> <4FF3E2C4.7050701@FreeBSD.org> <4FF3FB14.8020006@FreeBSD.org> <4FF402D1.4000505@FreeBSD.org> <17110267579.20120704233705@serebryakov.spb.ru>
next in thread | previous in thread | raw e-mail | index | archive | help
On 04.07.2012 23:37, Lev Serebryakov wrote: > Hello, Alexander. > You wrote 4 èþëÿ 2012 ã., 12:46:09: > > > AVC> http://shader.kaist.edu/packetshader/ (and links there) are good example > AVC> of what is going on. > But HOW?! GPU has very high "preparation" and data transfer cost, > how it could be used for such small packets of data, as 1.5-9K > datagrams?! According to http://www.ndsl.kaist.edu/~kyoungsoo/papers/packetshader.pdf - cumulative dispatch latency is between 3.8-4.1 microseconds (section 2.2). And GPU is doing routing lookup only (at least for IPv4/IPv6 forwarding), so we're always transferring/receving fixed amount of data. Btw, there are exact hardware specifications in this document. >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?4FF4A0A7.9050601>