Trevor Talbot skrev: > On Dec 17, 2007, at 8:59 PM, Iustin Pop wrote: > > >> On Mon, Dec 17, 2007 at 11:28:51PM +0100, Pontus wrote: >> > > >>> Oh, that sounds a bit discouraging. Isn't CF just a variant of the IDE >>> bus, and should therefore be able to reach speeds of 33 - 133 MB/s >>> depending on the controller. I realize that a caveat in the OpenBSD >>> driver could slow things down, but all the way to 2MB/s ? I'm not >>> saying >>> you are wrong, I'm just surprised. Have you done some benchmarks? >>> >> FWIW, on a 5501 with DMA enabled and on Linux, using a SanDisk >> SDCFX3-2048 (and indeed multi-sector): >> >> # dd if=/dev/hda1 of=/dev/null bs=128k iflag=direct >> 377+1 records in >> 377+1 records out >> 49512960 bytes (50 MB) copied, 6.80798 seconds, 7.3 MB/s >> > > Flash is not very fast in general. Spinning disks are much faster at > sequential reads. > > The performance gain you always hear about with flash is actually the > random access speed, because solid-state media doesn't have seek times. > (I'm sort of straying away from the original discussion now, tell me if I should mark this OT)
I've always thought that do be a bit odd, not that seek times are better, but that it speeds has to be slower just because it is flash. I understand that one single flash chip will have its speed limitation, but by grouping multiple chips together and reading from them in parallel(like a raid-0), that should give a big boost in performance. I guess that is what's done in the fusion-io card: http://www.fusionio.com/ (a little less droid speak: http://www.tgdaily.com/content/view/34065/135/) It is a bit pricey, to say the least, but fast. /Pontus. _______________________________________________ Soekris-tech mailing list [email protected] http://lists.soekris.com/mailman/listinfo/soekris-tech
