> -----Original Message----- > From: ceph-users [mailto:ceph-users-bounces@xxxxxxxxxxxxxx] On Behalf Of > GODIN Vincent (SILCA) > Sent: 07 May 2015 11:13 > To: ceph-users@xxxxxxxxxxxxxx > Subject: About Ceph Cache Tier parameters > > Hi, > > In Cache Tier parameters, there is nothing to tell the cache to flush dirty > objects on cold storage when the cache is under-utilized (as far as you 're > under the "cache_target_dirty_ratio", it's look like dirty objects can be > keeped in the cache for years). Yes this is correct, I have played around with a cron job to flush the dirty blocks when I know the cluster will be idle, this improves write performance for the next bunch of bursty writes. I think the idea behind the current cache thinking is more geared to something like running VM's where typically the same hot blocks will be written to over and over again. My workload involves a significant number of blocks which are written once and then never again and so flushing the cache before each job run seems to improve performance. > > That is to say that the flush operations will always start during writes and > when we have reached the "cache_target_dirty_ratio" value : this will slow > down the current writes IO. > > Are some futur requests planned to improve this behavior ? Not that I'm currently aware of, but I did post here a couple of weeks ago suggesting that maybe having high and low watermarks for the cache flushing might improve performance. At the low watermark, cache would be flushed with a low/idle priority (much like scrub options) and at the high watermark the current flushing behaviour would start. I didn't get any response, so I think this idea may have hit a bit of a dead end. I did start having a looking through the Ceph source to see if it was something I could try doing myself, but I haven't found enough time to get my head round it. > > Thanks for your response > > Vince > > > ________________________________________ > Ce message et toutes les pi?ces jointes (ci-apr?s le "Message") sont > strictement confidentiels et sont ?tablis ? l'attention exclusive de ses > destinataires. > Si vous recevez ce message par erreur, merci de le d?truire et d'en avertir > imm?diatement l'exp?diteur par e-mail. > Toute utilisation de ce message non conforme ? sa destination, toute > modification, ?dition, ou diffusion totale ou partielle non autoris?e est > interdite. SILCA d?cline toute responsabilit?; au titre de ce Message s'il a ?t? > alt?r?, d?form?, falsifi? ou encore ?dit? ou diffus? sans autorisation. > This mail message and attachments (the "Message") are confidential and > solely intended for the addressees. > If you receive this message in error, please delete it and immediately notify > the sender by e-mail. > Any use other than its intended purpose, review, retransmission, > dissemination, either whole or partial is prohibited except if formal approval > is granted. SILCA shall not be liable for the Message if altered, changed, > falsified , retransmeted or disseminated. _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
 |