Date: Thu, 02 Oct 2008 16:39:56 -0600 From: Scott Long <scottl@samsco.org> To: John Baldwin <jhb@freebsd.org> Cc: Joel Jacobson <jjacobson@panasas.com>, freebsd-current@freebsd.org Subject: Re: scatter/gather in struct buf/bio Message-ID: <48E54DBC.5030005@samsco.org> In-Reply-To: <200810021553.12976.jhb@freebsd.org> References: <2DC640DB-2B7C-4111-9842-BDFA3E134282@panasas.com> <200810021553.12976.jhb@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
John Baldwin wrote: > On Thursday 02 October 2008 02:11:38 pm Joel Jacobson wrote: >> Greetings - >> >> I have something which would be vastly improved by the ability to >> construct my own sgl to pass through a buf, and ran across: >> >> http://wiki.freebsd.org/Buf0x >> >> Is there somewhere I can follow progress, and/or contribute assistance? > > The jhb_bio branch still exists. I've made some refinements to the sglist API > and need to do so more. I think my next test will be to add a sglist to each > bio in geom_disk and geom_vfs. I then might try to update physio (dd > if=/dev/XXX) to not map buffers into KVA (probably tweakable via sysctl) and > patch ATA to use the new bus dma call with the sglist in the bio and see what > difference that makes. > Doing a single dd probably isn't going to show any difference at all. That won't mean that the changes aren't worthwhile, it'll just mean that you're testing the seek latency of the disk, which is a property that is reliably static. Please talk to me in private about effective ways to do performance testing for something like this. Scott
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?48E54DBC.5030005>