Hi, This is a valid point. Unfortunately, under some conditions, such as one client reading with 1 thread and low queue depth, the read performance leaves much to be desired. I'm wondering if there's a way to improve the performance without changing to much faster storage drives. Z On Thu, Nov 11, 2021 at 7:00 PM Patrick Donnelly <pdonnell@xxxxxxxxxx> wrote: > On Thu, Nov 11, 2021 at 7:55 AM Zakhar Kirpichenko <zakhar@xxxxxxxxx> > wrote: > > > > Hi, > > > > I'm still trying to combat really bad read performance from HDD-backed > > replicated pools, which is under 100 MB/s most of the time with 1 thread > > and QD=1. I don't quite understand why the reads are that slow, i.e. much > > slower than a single HDD, but do understand that Ceph clients read a PG > > from primary OSD only. > > > > Since reads are immutable, is it possible to make Ceph clients read PG > in a > > RAID1-like fashion, i.e. if a PG has a primary OSD and two replicas, is > it > > possible to read all 3 OSDs in parallel for a 3x performance gain? > > It sounds like a compelling idea but you have to consider read-write > ordering: you have the potential for a read on a replica returning > out-of-date data. There are strategies to resolve that but you have to > ask yourself why? RADOS specifically scales by adding more > objects/PGs. This should naturally spread load on primary OSDs across > the cluster. Replicas are for redundancy not for improving (read) > throughput. > > -- > Patrick Donnelly, Ph.D. > He / Him / His > Principal Software Engineer > Red Hat, Inc. > GPG: 19F28A586F808C2402351B93C3301A3E258DD79D > > _______________________________________________ ceph-users mailing list -- ceph-users@xxxxxxx To unsubscribe send an email to ceph-users-leave@xxxxxxx
 |