Re: Question about erasure code

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

 



Den tis 28 jan. 2020 kl 17:34 skrev Zorg <zorg@xxxxxxxxxxxx>:

> Hi,
>
> we are planning to use EC
>
> I have  3 questions about it
>
> 1 / what is the advantage of having more machines than (k + m)? We are
> planning to have 11 nodes and use k=8 and m=3. does it improved
> performance to have more node than K+M? of how many ? what ratio?
>

You should always try to have one or two (or many more!) hosts than the
replication
size or the sum of K+M in EC. If you run exactly with K+M hosts, any
surprise means the
cluster is degraded. One host falling over means that ALL pgs are now in
degraded mode and
most will resort to rebuilding data via checksums on read. This is bad.

If you ever plan to do maintenance on a host (be it OS upgrades, package
upgraded, the non-ending
Intel mitigation patches, whatever), and you don't have spare hosts then
you are getting into
degraded mode even for planned downtimes, as well as for all the unplanned
ones.

Having more nodes (regardless of if you do replication or EC), more hosts
bring more networking capacity, more CPU, more RAM for caches, more IOPS in
total and
more ways to share the load of normal usage and recovery, not just "getting
more space"


> 2 / what behavior should we expect if we lose 1,2 or 3 nodes: perfs /
> recovery. how many nodes could we loose before the cluster state change
> to read only
>

Depends a bit on the M of course, but EC in ceph will not be happy at K
hosts,
so K+1 is minimum for it to work and recover by itself.


-- 
May the most significant bit of your life be positive.
_______________________________________________
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