Hello, On Wed, 29 Oct 2014 23:01:55 +0200 Cristian Falcas wrote: > Hello, > > Will there be any benefit in making the journal the size of an entire > ssd disk? > Not really, Craig already pointed out a number of things. To put numbers on things, I size my journals so they can at least hold 20 seconds worth of writes (reasoning below) at full speed. For example a OSD server with a single 10Gb/s link, 8 OSDs and 4 journal SSDs (2 journals per SSD), each SSD capable of writing 250MB/s (or faster). So 20 seconds worth through that 10Gb/s link make 20GB, divided by 8 results in a mere 2.5GB journal. Now I still deployed machines like the above with 10GB journals because I had the space, but that's about it. > I was also thinking on increasing "journal max write entries" and > "journal queue max ops". > > But will it matter, or it will have the same effect as a 4gb journal > on the same ssd? > Find the "How to detect journal problems" thread in the ML archives from April this year for details. In short the journal* options don't necessarily do what you think they do, if you want to max out journal utilization you have to play with the filestore min and max sync intervals. I have set the max to 10 seconds and since Ceph also starts flushing the journal when it becomes half full there's the above goal of having 20 seconds worth of space. That all said, I'd be very happy about some journal perf counters in Ceph that show how effective and utilized it is. Christian -- Christian Balzer Network/Systems Engineer chibi@xxxxxxx Global OnLine Japan/Fusion Communications http://www.gol.com/ _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |