On Mon, 29 Apr 2013, James Harper wrote:
If anyone has a few seconds spare, could you please run the following and post the results:
dd if=/dev/zero of=test.bin bs=512 count=128 oflag=sync
along with the kernel version and arch, filesystem, layers (lvm, md, etc), and underlying hardware?
When I do it on a bare 7200RPM sata disk on a modern server running xfs I get 10-15kbytes/second. I've repeated this on 3 other servers on different hardware with similar results, but when I do it on a ~10yo PC with ext3 I get around 500kbytes/second - 50x faster. I suspect it might be the kernel version (3.8 on new server, 2.6 on old pc) and the implementation of O_SYNC in older kernels wrt metadata but I don't have enough data points to form any conclusions...
What a fun excercise. At work with you beaut SANs tier 1 storage with 15k SAS disks and SSD cache, still only getting 600k (some of our dev tier 3 stuff comes out quicker - too many VM datastores to disentangle how heavily loaded each datastore is though). About the same as my 5 year old laptop with a 240G SSD in it running 3.8. But about 60 times faster than my raspberry pi. Imagine a beowulf cluster of those. And my root filesystem and (empty) home directory on my ZFS fileserver boots off a USB nanostick, which gets the same performance as a ZFS pool on a real disk (er, the raidz 3 disk pool hasn't come back after a minute, but that disk is rather busy at this moment. That box never has been sane). -- Tim Connors