Here are plots that show the average aio latency for each commit block with different levels of concurrency generated by fio doing 4k random writes working with objectstore (bluestore). Concurrency is 16, 32, 64, and 128 as controlled by the iodepth parameter to fio. (I’m sending links to the plots rather than attaching them, as apparently messages with attachments never make it out to ceph-devel.) concurrency 16 http://tracker.ceph.com/attachments/download/2694/fio-b-aio-latency-t16.pdf concurrency 32 http://tracker.ceph.com/attachments/download/2695/fio-b-aio-latency-t32.pdf concurrency 64 http://tracker.ceph.com/attachments/download/2696/fio-b-aio-latency-t64.pdf concurrency 128 http://tracker.ceph.com/attachments/download/2697/fio-b-aio-latency-t128.pdf concurrencies overlaid http://tracker.ceph.com/attachments/download/2698/fio-b-aio-latency-tmulti.pdf For reference, here are the *previous* round of plots generated by rados bench, also doing 4k random writes, going through the full ceph stack. Also for each plot I tried fitting quadratic, cubic, and quartic polynomials to the data points, and I overlay those curves. concurrency 16 http://tracker.ceph.com/attachments/download/2699/regbench-l-t16-fit-nat.pdf concurrency 32 http://tracker.ceph.com/attachments/download/2700/regbench-l-t32-fit-nat.pdf concurrency 64 http://tracker.ceph.com/attachments/download/2701/regbench-l-t64-fit-nat.pdf concurrency 128 http://tracker.ceph.com/attachments/download/2702/regbench-l-t128-fit-nat.pdf concurrencies overlaid http://tracker.ceph.com/attachments/download/2703/regbench-l-tmulti-fit-nat.pdf Eric-- 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
