On Mon, Nov 29, 2010 at 10:19 PM, Jeff Wu <cpwu@xxxxxxxxxxxxx> wrote: > Is "40-50MB/s" the speed that it run bench at local btrfs disk ? > not the speed that run bench from client to osd server ? > with this speed ,run bench from client to osd server ,will which get > about 20~25MB/s( 40~50MB /2 )speed ? Data on Ceph is replicated across 2 OSDs (by default; this is configurable). So while figuring out potential performance involves a lot of variables, in a simple case like this where you aren't bounded by network bandwidth you'll find that your read/write performance simply tracks the slower disk. I'd expect your Ceph tests (at least the streaming ones) to run at 40-50MB/s. Given that everything else is okay, I cannot stress enough that running without a journal is going to cause significant performance degradations. I have a hard time believing that it's responsible for 13-second latencies, but it's possible. So how about you set up a journal (it can just be a file or new partition on the drives you're already using) and report back your results after you do that. :) Adding a journal to the OSDs lets them turn all their random writes into streaming ones. -Greg -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
