On Wednesday 13 July 2011, Manoj Iyer wrote: > > Chris/Arnd, > > Here is a series of test I did with the patched kernel. > == cold boot insert SD card == > u@u:~/flash/flashbench$ sudo ./flashbench -O --erasesize=$[4 * 1024 * > 1024] --blocksize=$[256 * 1024] /dev/mmcblk0 --open-au-nr=2 > 4MiB 4.96M/s > 2MiB 6.3M/s > 1MiB 6.23M/s > 512KiB 6.23M/s > 256KiB 6.26M/s The very first one obviously triggers a garbage collection. Everything after that is well within measuring accuracy around 6.25MB/s > On Wed, 13 Jul 2011, Chris Ball wrote: > > > > That makes sense. Do you think this explains Manoj getting a slower > > first file copy speed (757ms vs. 480ms) after applying his patch? > > (Manoj, perhaps you could retry your test without GC being needed?) Yes. For a single sample, it can easily explain differences up to 500ms. You have to average out file system benchmarks across a lot of files to be sure. > > What would we expect lowering the SD base clock frequency from 200MHz > > to 50MHz to do to performance theoretically? Not much. This card only has a 6MB/s write speed, which is well below what a 50 MHz bus can do. It mgiht be different on a fast eMMC device or a Sandisk Extreme Pro UHS-1 card. Arnd -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
