Best practice on disk setup

Discussion in 'Pre-Deployment: Getting Started and Installation' started by RobertOe, Sep 3, 2014.

  1. RobertOe

    RobertOe Bit Poster

    Hi guys

    I'm having 3 new servers are about to set them up as shared MDS/CS/Client . However, I get a bit confused about the best practice on setting up the storage.

    We based our servers on the PDF document called "Best Practices for Configuring a Parallels Cloud Storage Cluster". It states that we should use RAID0 on the SAS disk, but looking at the instalastion manual it recommends to not use RAID0. Would it be best to have all the SAS disks in one RAID0, or leave them as single disks and that way have more clusters and less things to go to "hell" if one disk dies? I guess leaving one disk at least for the OS would be a good idea. We started "low" and have only 6 X 2TB disks on each server.

    In additioin, we have as the document recommended, 2 X 100GB SSD disks. Would it be advisable to split the work so that one disk handles the write-cache, while the other disk handles the read-cache? Or should both disk handle both read-cache and write-cache set up like in the innstalation-manual in the chapter about using SSD disks. (ex. 20% checksum, 40% read, 40% write). Will the server be able to use both disks that way, or would I need to use a raid0 on them in that setting?

    Hope someone got some hints for me so I can get our new servers set up :)

    Robert Oedegaard
  2. akirov

    akirov Bit Poster

    Hi Robert,

    You should use independent HDDs – each HDD is a dedicated RAID0 volume (the volume of one disk) or individual disk in JBOD. “One HDD = one volume = one CS”.

    You should use two independent SSD without union in RAID0. Note, that usually it’s better to turn off RAID cache for SSD disks, because modern SSD can work with performance better than RAID cache.
    For right SSD cache configuring in PStorage you can use "advise-configuration" command, as described in
    In your case, the command will be something like this:
    pstorage -c cluster_name advise-configuration --cs /hdd1 --cs /hdd2 --cs /hdd3 --cs /hdd4 --cs /hdd5 --cs /hdd6 --ssd /ssd1 --ssd /ssd2 -m /pstorage/cluster_name
    PStorage will calculate cache sizes, based on your disks set, and will advise commands to configure it.

    PS: By the way, we recommend to use in production replica=3 and at least 5 physical hosts.

    Alexander Kirov
  3. Machiasiaweb

    Machiasiaweb Mega Poster


    I want to further ask some questions. Is that have concept of read/write cache? or Just a single file?

    Also repeat same question for Journal as well?

    Further question is if it put cache and Journals into different SSD disk. Will the whole platform performance will increase?


Share This Page