Re: How much space?

JuanJose Galvez <juanjose.galvez@xxxxxxxxxxx> · Sun, 22 Dec 2013 15:10:29 -0800



    On 12/22/2013 1:57 AM, shacky wrote:

    
              Replication
                  is set on a per pool basis. You can set some, or all,
                  pools to replica size of 2 instead of 3.
            
            
            Thank you very much. I saw this is to be setted in the
              global configuration (osd pool default size).
            So it's up to me to configure Ceph to be rendundant and
              fault tolerant?
          
        
    The default size is already 2, so it will be redundant and somewhat
    fault tolerant by default. You can learn quite a bit more about
    pools by reviewing the information in the docs here:

    
    http://ceph.com/docs/master/rados/operations/pools/

    
    Just know that replication does not have to be the same across all
    pools.

    
            If I set "osd pool default size" to 2, I will be sure
              that if a cluster node goes down my data will be safe?
          
        
    That is just the default value. Another default is replication being
    done so that we never place a copy of the data on the same host.
    Depending on your infrastructure you may want to expand this to
    racks or rows. In any case losing a single node generally means that
    your data will be safe (according to the defaults). The potential
    for data loss would be there if you lost two drives in different
    hosts BEFORE the cluster finished recovery from the first drive. The
    documentation regarding the CRUSH map is your friend when it comes
    to understanding all of this:

    
    http://ceph.com/docs/master/rados/operations/crush-map/

    
              Ceph
                  uses replication not erasure coding (unlike RAID). So
                  data is completely duplicated in multiple copies.
                  Erasure coding is scheduled for the Firefly release,
                  according to the roadmap.
            
            
            Ad I said, I expect having something similar to RAID5:
              if one hard drive per cluster node fails, my data will be
              safe. If an entire cluster node fails, my data will be
              safe. Could you help me to understand the correct
              configuration for this situation?
          
        
    Losing an entire cluster node, by default, your data would be safe.
    However if you were to lose one drive on each node, quickly enough
    (before recovery completes), it would be possible to have lost some
    data. To avoid that you could setup RAID sets behind each OSD but
    that will drive up your cost per gigabyte and, depending on the RAID
    configuration, could mean having to replicate larger amounts of data
    when you do lose an OSD. Also, that sort of setup could have
    performance implications involved.

    
    I prefer to see drives setup in JBOD, pools with a replication level
    3, a properly implemented infrastructure in the CRUSH map, all
    sitting on (at least) a 10Gbps cluster network.

    
    It might be a bit daunting at first, as there is a lot to learn when
    it comes to Ceph, but the documentation really is going to be worth
    the read. From your questions I would suggest going through the
    Architecture documentation which explains RADOS and how data is
    stored. Understanding how Ceph stores data will give you a better
    idea of how replication and failures are handled.

    
    http://ceph.com/docs/master/architecture/

    
            Thank you very much for your help!
            Bye.
          
        
    Good luck!

    -- 
JuanJose "JJ" Galvez
Professional Services
Inktank Storage, Inc.
LinkedIn: http://www.linkedin.com/in/jjgalvez
  

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com