Re: [ceph-users] Sizing SSD's for ceph
Hi, Am 29.01.2015 07:53, schrieb Christian Balzer: On Thu, 29 Jan 2015 01:30:41 + Ramakrishna Nishtala (rnishtal) wrote: * Per my understanding once writes are complete to journal then it is read again from the journal before writing to data disk. Does this mean, we have to do, not just sync/async writes but also reads ( random/seq ? ) in order to correctly size them? You might want to read this thread: https://www.mail-archive.com/ceph-devel@vger.kernel.org/msg12952.html Assuming this didn't change (and just looking at my journal SSDs and OSD HDDs with atop I don't think so) your writes go to the HDDs pretty much in parallel. In either case, an SSD that can _write_ fast enough to satisfy your needs will definitely have no problems reading fast enough. due, that the data are in the cache (ram), there are only marginal reads from the journal-ssd! iostat from an journal ssd: Device:tpskB_read/skB_wrtn/skB_readkB_wrtn sdc 304,45 0,16 82750,46 29544 15518960008 I would say, if you have much more reads, you have to less memory. Udo ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
Re: [ceph-users] Sizing SSD's for ceph
On Thu, 29 Jan 2015 01:30:41 + Ramakrishna Nishtala (rnishtal) wrote: Hi, Apologize if something came up before like this. Reading archives, it appears that 4 to 5 spinning disks are recommended for single SSD. It all depends on the SSDs and HDDs in question for one (how many HDDs can that single SSD keep busy) and on the impact you're willing to take in case of a SSD failure. I have two questions on the subject. * Some of the links suggest that we should use 'sync writes' to really size the journals. If true, then what is journal_aio = true means? Does ceph use async IO for journals with this parameter and sync write is not a true test for sizing journals? No idea on that one. * Per my understanding once writes are complete to journal then it is read again from the journal before writing to data disk. Does this mean, we have to do, not just sync/async writes but also reads ( random/seq ? ) in order to correctly size them? You might want to read this thread: https://www.mail-archive.com/ceph-devel@vger.kernel.org/msg12952.html Assuming this didn't change (and just looking at my journal SSDs and OSD HDDs with atop I don't think so) your writes go to the HDDs pretty much in parallel. In either case, an SSD that can _write_ fast enough to satisfy your needs will definitely have no problems reading fast enough. Christian -- Christian BalzerNetwork/Systems Engineer ch...@gol.com Global OnLine Japan/Fusion Communications http://www.gol.com/ ___ ceph-users mailing list ceph-users@lists.ceph.com http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com