Date: Thu, 2 Oct 2003 14:12:59 +1000 (EST) From: Bruce Evans <bde@zeta.org.au> To: Nate Lawson <nate@root.org> Cc: current@freebsd.org Subject: Re: umass(4)/uhci(4) REALLY slow Message-ID: <20031002135340.O8330@gamplex.bde.org> In-Reply-To: <20031001111502.F85421@root.org> References: <20030930172903.S82394@root.org> <20031001142825.K4031@gamplex.bde.org> <20031001111502.F85421@root.org>
index | next in thread | previous in thread | raw e-mail
On Wed, 1 Oct 2003, Nate Lawson wrote: > On Wed, 1 Oct 2003, Bruce Evans wrote: > > On Tue, 30 Sep 2003, Nate Lawson wrote: > > > Here are "iostat 5" results for my USB thumb drive on a uhci(4) controller > > > with 5.1-CURRENT. On windows on the same box, it runs reasonably quickly. > > > On FreeBSD, it really lags. This is for a cp of a large file to a > > ... > > This is probably due to something we're not doing in msdosfs. 1K is > > probably your msdosfs file system block size. > > Yes, I checked and it has a 1K block size. The flash device is 64 MB. > > > msdosfs is missing support > > for clustering. None of the lower levels (buffer cache, driver, usb) > > in FreeBSD does clustering (the buffer cache has some support for it, > > ... > > What would need to be done to add msdosfs clustered reads/writes or > perhaps do this in a more general way in the buffer cache? Not a lot for msdosfs. Basically, add some B_CLUSTEROK's and vfs_bio_awrite()s, and 1 cluster_write() to msdosfs_write(). Merge them from ffs_write(). Better, merge more of ffs_write(). IIRC, using cluster_write() is only a small optimization -- write clustering mostly work if you throw everything into the buffer cache using bdwrite() without neglecting to set B_CLUSTEROK. Not much more for the buffer cache if you only do what bdwrite() and B_CLUSTEROK do. Non-delayed writes can't be clustered very effectively at the buffer cache level. Similarly for read clustering except it is cluster_read() and larger read-ahead instead of cluster_write() and B_CLUSTEROK/vfs_bio_awrite() that are needed. The file system must be more involved since reads are less predictable than writes at the buffer cache level. Brucehome | help
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20031002135340.O8330>