Re: Wide variation in osd_mclock_max_capacity_iops_hdd

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Yes, device caching is disabled. The ODSs in question use a separate DB/WAL device on flash. Could that be the cause of IOPS over-estimation?

Is it better for me to manually set osd_mclock_max_capacity_iops_hdd to a reasonable number or will things perform OK as long as IOPS are over-estimated consistently everywhere?


Vlad



On 9/6/22 11:50, David Orman wrote:
Yes. Rotational drives can generally do 100-200IOPS (some outliers, of course). Do you have all forms of caching disabled on your storage controllers/disks?

On Tue, Sep 6, 2022 at 11:32 AM Vladimir Brik <vladimir.brik@xxxxxxxxxxxxxxxx <mailto:vladimir.brik@xxxxxxxxxxxxxxxx>> wrote:

    Setting osd_mclock_force_run_benchmark_on_init to true and
    restarting OSDs fixed the problem of high variability.

    However, osd_mclock_max_capacity_iops_hdd is now over 2000.
    That's way too much for spinning disks, isn't it?

    Vlad



    On 9/1/22 03:54, Sridhar Seshasayee wrote:
     > Hello Vladimir,
     >
     >     I have noticed that our
    osd_mclock_max_capacity_iops_hdd
     >     varies widely for OSDs on identical drives in
    identical
     >     machines (from ~600 to ~2800).
     >
     > The IOPS shouldn't vary widely if the drives are of
    similar
     > age and running
     > the same workloads. The
    osd_mclock_max_capacity_iops_hdd is
     > determined
     > using Ceph's osd bench during osd boot-up. From our
    testing
     > on HDDs, the
     > tool shows fairly consistent numbers (+/- a few 10s
    of IOPS)
     > for a given HDD.
     > For more details please see:
     >
    https://docs.ceph.com/en/latest/rados/configuration/mclock-config-ref/#osd-capacity-determination-automated <https://docs.ceph.com/en/latest/rados/configuration/mclock-config-ref/#osd-capacity-determination-automated> <https://docs.ceph.com/en/latest/rados/configuration/mclock-config-ref/#osd-capacity-determination-automated <https://docs.ceph.com/en/latest/rados/configuration/mclock-config-ref/#osd-capacity-determination-automated>>
     >
     > In your case, it would be good to take another look
    at the
     > subset of HDDs
     > showing degraded/lower IOPS performance and check if
    there
     > are any
     > underlying issues. You could run the osd bench
    against the
     > affected osds as
     > described in the link below or your preferred tool to
    get a
     > better understanding:
     >
     >
    https://docs.ceph.com/en/latest/rados/configuration/mclock-config-ref/#benchmarking-test-steps-using-osd-bench <https://docs.ceph.com/en/latest/rados/configuration/mclock-config-ref/#benchmarking-test-steps-using-osd-bench> <https://docs.ceph.com/en/latest/rados/configuration/mclock-config-ref/#benchmarking-test-steps-using-osd-bench <https://docs.ceph.com/en/latest/rados/configuration/mclock-config-ref/#benchmarking-test-steps-using-osd-bench>>
     >
     >     Is such great variation a problem? What effect on
     >     performance does this have?
     >
     > The mClock profiles use the IOPS capacity of each osd to
     > allocate IOPS
     > resources to ops like client, recovery, scrubs and so
    on.
     > Therefore, a lower IOPS
     > capacity will have an impact on these ops and
    therefore it
     > would make sense to
     > check the health of the HDDs that are showing lower than
     > expected IOPS numbers.
     > -Sridhar
     >
    _______________________________________________
    ceph-users mailing list -- ceph-users@xxxxxxx
    <mailto:ceph-users@xxxxxxx>
    To unsubscribe send an email to ceph-users-leave@xxxxxxx
    <mailto:ceph-users-leave@xxxxxxx>

_______________________________________________
ceph-users mailing list -- ceph-users@xxxxxxx
To unsubscribe send an email to ceph-users-leave@xxxxxxx




[Index of Archives]     [Information on CEPH]     [Linux Filesystem Development]     [Ceph Development]     [Ceph Large]     [Ceph Dev]     [Linux USB Development]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [xfs]


  Powered by Linux