pros/cons of many single drives (jbod) or raid, and how the sw. client react to a failure



  • question, i'd rather not buy a raid card for utilizing some spare 4TB drives I have, and instead add them as new F:, G:, H:, I: drives

    Assuming the mining client can have 4 plots and mine against the 4 different disks? which mining client does this the easiest/best?

    i saw n the videos the plot setup and it seems i'd just need to 'replot' 4 times if i have 4 separate disks, not the end of the world if it's only done that once. Going with a raid card to make a big ole single volume would avoid this small task, so that alone isn't worth it, any other reason(s) to have fewer/larger volumes by either using a raid card or simply doing a stripe/extended disk (spanning multiple physical drives into one large logical volume which showed up as a single/big ~8TB F: drive?)

    which clients support GPUs natively to assist with plotting? delta in time to plot an 8TB volume/drive with normal quad core cpu vs. current gen gpu?

    thanks,

    nobody



  • I see a post by blago in another thread mentioning having > 1 volume allowing for parallelization:

    "...
    @issaburst if you have plots at two drives (for ex. d:\plots\ and e:\plots), in miner's config you may set (parallely reading)

    "Paths":["D:\plots","E:\plots"],
    ..."

    and samples of config files where having individual drives addressed by the OS seems more popular. also the best for actual results? or just more common/easier for the avg user?



  • @nobody if you make RAID - you lost some time for reading plots (access to file).
    better choice: 1 drive = 1 optimized plot, no RAID



  • ok, so no raid. just individual drives.

    about how many files per TB does the plot produce? (what size are the files?)

    and how many total files can one windows host manage without difficulty (ie: if I have some jbod shelves, I can SAS connect several to an external sas hba, so dozens of drives all from one windows server 'head')

    assuming server is 8 cores and 24GB ram, OS and app is on ssd drive, plot storage is 7200rpm sata via sas hba



  • also (sorry) what's the impact to the process if(when) a drive dies?

    since we have say 4 plot drives, each with unique data on it, and no raid-parity, is it simply a matter of any data on that drive 'never gets processed and turned in' so other than losing productivity, there's no overall larger issue?



  • @nobody RAM volume does not really matter.
    If you loose 1 drive of 4 - miner will try to get access to it, but other still works fine and will mining at 3 drives.
    http://imgur.com/a/0llBI my M:\ drive is dead yesterday

    If you have many plots at 1 drive - you will loose 8ms per each for get access to it