PCS - HA

Discussion in 'Installation, Update and Configuration' started by futureweb, Feb 18, 2016.

  1. futureweb

    futureweb Tera Poster

    Messages:
    397
    Hello,
    one Question about HA Mode with Virtuozzo Storage. As of install manual - Prerequisite for HA Mode is "a cluster has 5 or more chunk servers".
    Now I wanted to ask --> Why 5?

    As I understand 3 should be enough as it would still have Quorum if one Server goes down?
    I also just set up a Test-Case with 3 Servers and it's working finde ...

    Could someone please give me some insight why 5 CS is listed as a Prerequisite?

    Thx
    Andreas Schnederle-Wagner
     
  2. KonstantinB

    KonstantinB Odin Team

    Messages:
    68
    Hello,

    Quorum is about MDS servers.
    5 CSes are needed to ensure that all data is available for both read and write in case of node failure.
    With 3 CSes you will not loss your data, but there is high probability that write operations will be blocked until third CS become ready.
     
  3. futureweb

    futureweb Tera Poster

    Messages:
    397
    Hello Konstantin,

    could you please tell me under which circumstances write operations will be blocked in such a 3 CS Scanario?
    As we are planing our new Hardware Setup right now I try to figure out the way we gonna go ... (and it's quite a pricy difference if one goes for 3 Servers or 5 Servers)

    Thx
    Andreas
     
  4. KonstantinB

    KonstantinB Odin Team

    Messages:
    68
    Hello Andreas,

    Key issue here is that data chunk states are dynamic. In static case, 3 servers are sufficient. But in dynamic...
    Please review next documentation chapter - http://updates.virtuozzo.com/doc/pc..._Cloud_Storage_Administrators_Guide/36355.htm
    For example, if you have few chunks in 'Replicating' state and one node crashed all future writes to these chunks will be blocked. And with only 2 CSs left replication will never complete.

    Same for 'Pending' and even worse for 'Urgent' and 'Degraded' in case of previous short outages.
    Even during normal operation, one CS could be unavailable for say 5 seconds which could move chunks from stable 'Healthy' state.

    You could use storage cluster with only 3 nodes. This is reasonable configuration. Just don't treat it as High Availability and don't setup it there.

    Hope my answer help...
     
  5. futureweb

    futureweb Tera Poster

    Messages:
    397
    Hello Konstantin,

    for proper HA setup - should all 5 CS have the same amount of Storage or is mixed ok too?
    mixed like:
    - Node 1-3: 2TB
    - Node 4-5: 1TB

    Or is then 1TB HA ready and the rest not?! ;-)

    Andreas
     
  6. KonstantinB

    KonstantinB Odin Team

    Messages:
    68
    Hello Andreas,

    Yes, this setup is HA ready. Just don't fill storage up to the edge.
    If node with 2TB disk fails you need enough space on other nodes to restore replication factor.

    Best regards,
     
  7. futureweb

    futureweb Tera Poster

    Messages:
    397
    alright - got it! :)
     
  8. futureweb

    futureweb Tera Poster

    Messages:
    397
    alright ... seems like we go for the 5-Node configuration ...
    3 very strong Servers (2x E5-2650 v3 10C, 256GB RAM, ...) and 2 slightly weaker Servers for MDS + CS
    What kind of CPU (Ghz, Cores, 1 or 2 Sockets) and how much RAM do you suggest for a Scenario like ours for those 2 additional Servers? (so they won't slow down the All-SSD Solution)
    Key data: 5 Nodes, 10 x 2TB Enterprise SSDs - 2 SSDs per Node - expandable to 12 SSDs per Node
     
  9. Pavel

    Pavel A.I. Auto-Responder Staff Member

    Messages:
    478
    Hi,

    If you plan to expand up to 12 SSD afterwards make sure your disk controller can handle this enormous amount of I\O. Occasionally we face situations when controller becomes a bottleneck.

    As for the amount of RAM - pstorage services themselves do not consume much resources, you might refer to the very same documentation and page 12 :)
    Thus, you may plan RAM and CPU based on VEs you'll run there.
     
  10. futureweb

    futureweb Tera Poster

    Messages:
    397
    alright - just wanted to make sure that the "minimum requirements" are also enough for SSD ONLY pstorage ... ;-)
    Controller will be HP P246br - guess it should be able to handle the Load.

    As of Documentation - Page 12:
    I guess TB should be GB ... right?
     
  11. Pavel

    Pavel A.I. Auto-Responder Staff Member

    Messages:
    478
    Nope, everything is correct here in terms of Tb/Gb sizes.
    MDS is a quite light-weight service.
     
  12. futureweb

    futureweb Tera Poster

    Messages:
    397
    well ... ok ... so I guess Xeon E5-2620 v3 with 32GB RAM + 2 x 2TB SSD should work as additional/dedicated MDS/CS Servers ... ;-)
     
  13. Pavel

    Pavel A.I. Auto-Responder Staff Member

    Messages:
    478
    If you're not going to run containers/vms over there that's more than enough, indeed.
     

Share This Page