EC does not lead to a performance degradation. EC and REP simply have different performance characteristics and, depending on your use case, one is better than the other. I did extensive testing before deciding what to go for and a general observation is, that on HDDs, EC has very high throughput while REP has good IOPs. With SSDs backing the pools, results are not so clear. Here, the latency of ceph's IO stack starts to dominate the results. This, in torn, heavily depends on the SSD models. Before you decide anything, here some points to consider: - Are you talking aggregated or single-thread performance? - What is the use case, large sequential IO or small random access? - Is if full-object access only or are objects going to be modified? - What are the backing devices, HDD or SSD? - How are the OSDs deployed, with fast WAL+DB/dm_cache or collocated? It is a huge parameter space and it changes from version to version, which is the reason why no data is published on the ceph docs. I explored only a small portion of this parameter space and spent 2 months full time on benchmarks. You will have to make some a-priori decisions to reduce the possibilities and then someone might be able to give details about a specific set up. Best regards, ================= Frank Schilder AIT Risø Campus Bygning 109, rum S14 ________________________________________ From: zp_8483 <zp_8483@xxxxxxx> Sent: 08 May 2021 10:45:17 To: ceph-users@xxxxxxx Subject: Performance compare between CEPH multi replica and EC Hi all, How much EC performance will degrade compared to multi replica when using the same hardware configuration. Is there any offical data? _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |