SSD for database server?

Discussion in 'Guides' started by jang430, May 30, 2018.

  1. jang430

    jang430 Member

    Joined:
    Mar 16, 2017
    Messages:
    79
    Likes Received:
    1
    Hi! I hope I'm posting on the right forum. I gather with all the server being discussed here, There are a lot of experts around :D

    I have an HP Proliant DL380p Gen8 server. It's missing a few drives. Is SSD (Samsung Evo, consumer) good enough to serve as it's hdd? RAID 5 will be implemented.

    I have the option to buy SAS drives that are specific to this old model. I reckon it will be more expensive due to it being old parts already. Consumer SSDs, though may not be as sturdy, nevertheless, has no moving parts. Can it do the job?

    Of course, regular backup will still be performed. This question is specifically for consumer SSD, such as Intel, or Samsung Evo only. I don't have access to used Enterprise models, as I see posted here a lot.
     
    #1
  2. JSchuricht

    JSchuricht Active Member

    Joined:
    Apr 4, 2011
    Messages:
    119
    Likes Received:
    35
    It all depends on what you are going to do with the system. Not all SSD's are equal. Most consumer SSD's are not optimized to handle a large amount of writes and will choke in some workloads. If you are in a lab environment or plan on keeping a small database in RAM with minimal flushing to the SSD it may be fine. If you have some mission critical workload with lots of writes to the SSD then you get what you pay for applies.
     
    #2
  3. BackupProphet

    BackupProphet Well-Known Member

    Joined:
    Jul 2, 2014
    Messages:
    720
    Likes Received:
    253
    For DB workloads, an enterprise class SSD can be up to 10x to 100x faster. Some consumer SSD's are almost as slow as SAS 10k spinners.
     
    #3
  4. jang430

    jang430 Member

    Joined:
    Mar 16, 2017
    Messages:
    79
    Likes Received:
    1
    I'm selling off the server. User wants SAS. I don't want them to get shortchanged, and yet, want lowest additional cost to add to the server before selling it.

    If performance won't be better, at least I need it to be at par, should be dependable too. Will consumer SSD such as Samsung Evo at least offer me that?
     
    #4
  5. gea

    gea Well-Known Member

    Joined:
    Dec 31, 2010
    Messages:
    1,885
    Likes Received:
    626
    For databases you want not only high write iops but also a transactional save write behaviour. So with old filesystems you need a hardwareraid + cache + BBU/Flash. With a modern filesystem like ZFS you want powerloss protection or an Slog that adds powerloss protection or sync performance to a pool that lacks this.

    Desktop SSDs do not offer high write iops under load or powerloss protection so no, I would not use Desktop SSDs and look more for a cheap ones from Intel DC series.

    SAS disks are not nearly comparable. They offer around 100 write iops while enterprise SSDs can go up to 50k write iops, NVMe up to 200k write iops and Intel Optane up to 500k.
     
    #5
  6. Evan

    Evan Well-Known Member

    Joined:
    Jan 6, 2016
    Messages:
    2,302
    Likes Received:
    319
    Given your selling it and being a HP server I wound strongly advise just getting used SATA or SAS HPE disks even if write intensive versions. Even the worst of them will out perform and work better that any Samsung EVO drive. Also with no risk of fans ramping up due to not compatible devices detected.

    Any 2.5” gen8 / gen9 smart carrier HPE SSD should essentially work.
     
    #6
  7. jang430

    jang430 Member

    Joined:
    Mar 16, 2017
    Messages:
    79
    Likes Received:
    1
    @Evan , Thanks. Used drives won't do with this customer. But will take your advice on the SAS HPE.
    @gea , thanks. You're saying SAS is the lowest, then Enterprise SSD, then NVMe SSD, and finally Intel Optane. Don't even consider desktop SSD, correct?
     
    #7
  8. jang430

    jang430 Member

    Joined:
    Mar 16, 2017
    Messages:
    79
    Likes Received:
    1
    @gea , BTW, so if I do get Intel DC series used drives, do I still need to worry about "hardwareraid + cache + BBU/Flash"? The unit does have hardware raid though. But I'm not sure about the cache, BBU/ Flash. I'm not familiar with them either. Will cache be with the hardware raid controller? What about BBU/ Flash? Do you need to enable them via BIOS? Or do special drivers have to be installed to make sure cache, BBU/ Flash will be used when powerloss occurs?
     
    #8
  9. gigatexal

    gigatexal I'm here to learn

    Joined:
    Nov 25, 2012
    Messages:
    2,481
    Likes Received:
    440
    What’s your/your clients budget?
     
    #9
  10. jang430

    jang430 Member

    Joined:
    Mar 16, 2017
    Messages:
    79
    Likes Received:
    1
    Very low. As for me, whatever way I can get the old servers out of the way, is what I'm after.
     
    #10
  11. gea

    gea Well-Known Member

    Joined:
    Dec 31, 2010
    Messages:
    1,885
    Likes Received:
    626
    The problem is called write hole problem, "Write hole" phenomenon in RAID5, RAID6, RAID1, and other arrays.

    in short
    If you write data to disk with a conventional filesystem, it first updates data then metadata entries. A crash between results in a corrupt filesystem. In a Raid ex Raid-1, the raidcontroller updates the first disk then the second. A crash between results in a corrupt raid (both parts of a mirror are different). In all of these cases none of the problems can be really fixed or detected. Only a hardeare raid controller with a cache + BBU/Flash protection can help (a little) as it can redo all commited writes on next reboot.

    A modern filesystem like ZFS offers CopyOnWrite to protect atomic writes (data+metadata is both written or none of them) and data + metadata checksums are there to decide if data is valid or not (example data on one part of a mirror or the other). This is why you do not need or want a hardware raidcontroller + cache protection on ZFS. But even ZFS must trust that a commit from disk means that data is really on disk. This is why - especially on SSDs you should care about SSDs with powerloss protection. With old file systems even in case of a hardware raidcontroller + cache protection.
     
    #11
  12. mbello

    mbello New Member

    Joined:
    Nov 14, 2016
    Messages:
    15
    Likes Received:
    7
    A very large number of databases used in companies are just a few gigabytes big.
    When that is the case, it becomes very cheap to get the best disk, you could go for 3 small Optane SSDs in Raid1.

    Else, for sure a consumer grade SSD will be much faster than the best SAS spinning disk out there. And yes, Samsung EVOs are quite reliable, just run it in proper RAID.

    But first do check how many GB your client really needs, 99% of the companies in this world do not have more than 100k customers and a few million orders per year. A database to handle that does not take much in disk space.
     
    #12
Similar Threads: database server
Forum Title Date
Guides How to rack a server Oct 29, 2018
Guides Create Custom Fan Speed Maps for SuperMicro X10 Servers via IPMI Apr 28, 2018
Guides GIGABYTE EL-20 Debian 8.5 Minimal Server install guide Aug 31, 2016

Share This Page