On Mon, 02 Mar 2009 17:19:22 +1300 (NZDT) John Carter <john.car...@tait.co.nz> wrote:
> On Mon, 2 Mar 2009, Craig Falconer wrote: > > > Well this is bizarre.... > > > > time dd if=/dev/zero of=/file.dd bs=16k count=16k > > produces a 256 MB file. > > > > onto sda (4GB) that took 11.34 seconds (27.9 MB/sec) > > However onto sdb it took 55.93 seconds (4.8 MB/sec) > > So you might be onto something here. > > > > time dd if=/dev/zero of=/file.dd bs=8k count=64k > > sda 39.2 sec 13.7 MB/sec > > sdb 77.2 sec 6.9 MB/sec > > > > Weird. > > Again you're probably testing the speed of ram, not flash. > > You need a test something like... > > sync;time bash -c 'dd if=/dev/zero of=/file.dd bs=8k count=64k;sync' > sync;time bash -c 'dd if=/dev/zero of=/file.dd bs=8k count=64k;sync' > sync;time bash -c 'dd if=/dev/zero of=/file.dd bs=8k count=64k;sync' > sync;time bash -c 'dd if=/dev/zero of=/file.dd bs=8k count=64k;sync' > > The sync to flush any dirty buffers, then the time 'bash -c 'dd > blah;sync' to time the dd AND the sync. Otherwise dd has just written > into ram buffers, which the kernel will flush at leisure. Noo... you should always sync twice: disk then network (: ( and anyone who says 3 times is even older than me! ) > > You also need to do it four times. Once to get the "bash", time and > sync "hot and in cache", three times to check you are getting > variability from other stuff happening. > > Bench marking on modern systems is tricksy in extreme. > > Usually I find cache effects dominate. > * Does it fit within L1, or L2 or disk buffers or not? > > * Is it in some level of cache already? > > * Is something squeezing it out of some level of cache? eg. A leaky > humongous javascript app running in your web browser whilst you > testing. (But nothing was running on my machine! A yes there was, > every minute or two the web server spat a new fat flash and > javascript ad at you!) > > Remember, Linux tries _hard_ to make use of _all_ your Ram _all_ the > time. It doesn't let it sit there idle and without value. > > John Carter Phone : (64)(3) 358 6639 > Tait Electronics Fax : (64)(3) 359 4632 > PO Box 1645 Christchurch Email : john.car...@tait.co.nz > New Zealand > The best one I ever came across was a non-technical ex boss who got a bunch of Indians ( schoolkids I think ) to benchmark some code. After starting off 4 copies of the code, the test failed, and a serious problem was indicated because they finished in a different order. A serious case of YGWYPF. Steve -- Steve Holdoway <st...@greengecko.co.nz>