SLOG drives

Discussion in 'Solaris, Nexenta, OpenIndiana, and napp-it' started by epicurean, Jul 31, 2018.

  1. epicurean

    epicurean Member

    Joined:
    Sep 29, 2014
    Messages:
    543
    Likes Received:
    20
    I have a couple of 100GB HGST HUSSL4010BSS600 drives that I am thinking of using as SLOG. This is for a pool of 20TB spinners in Z2 configuration under Napp It. Are they appropriate ? Will they make a difference
     
    #1
  2. pricklypunter

    pricklypunter Well-Known Member

    Joined:
    Nov 10, 2015
    Messages:
    1,514
    Likes Received:
    431
    Providing the SSD's have appreciably lower latency than your storage pool, then yes it'll work, if not, you'll be disappointed :)
     
    #2
  3. epicurean

    epicurean Member

    Joined:
    Sep 29, 2014
    Messages:
    543
    Likes Received:
    20
    How do I check their appropriate latency?
     
    #3
  4. i386

    i386 Well-Known Member

    Joined:
    Mar 18, 2016
    Messages:
    1,659
    Likes Received:
    400
    #4
  5. pricklypunter

    pricklypunter Well-Known Member

    Joined:
    Nov 10, 2015
    Messages:
    1,514
    Likes Received:
    431
    +1 Yes, definitely read the spec sheet for the drive in question and also run the numbers against your existing storage pool and see how it performs. Then once you have the facts before you, make a determination of what sort of gain/ return you can expect, if any :)
     
    #5
  6. gea

    gea Well-Known Member

    Joined:
    Dec 31, 2010
    Messages:
    2,213
    Likes Received:
    721
    The main question is whether you need or want sync writes.

    For an SMB filer you normally do not use or want sync write due the negative performance impact. The ZFS filesystem itself remains always valid due CopyOnWrite and the extra security to protect the content of the rambased writecache on a crash is quite small.

    This is different to a database use or when you use storage for VMs with older filesystems. A crash there can mean an inconsistent database ex when a financial transaction that consists from a remove money from one account and add it to another is only processed in its first part or on an older filesystem data is updated but not metadata resulting in a corrupt filesystem. To avoid this effects you definitely want and need sync write to protect the content of the writecache on a crash.

    A ZFS pool from one Z2 vdev made from disks has around 100 raw iops (without help of caches). Your HGST SSDs are rated 57K / 25K IOPS Random Read /Write so they are at least 200x better regarding iops. So yes if you use sync write they will help to improve write performance up to 10x say for example from 30MB/s to 300 MB/s compared to a unsync sequential write performance of for example 1000 MB/s with some disks.

    Another question would be if the HGST makes a good Slog?
    The answer is definitely yes if you compare against other Flash based SSDs. It is made of fast SLC with powerloss protection. An Intel DC 3700 (mostly regarded as best Sata option) may be a little faster but they are both the upperclass Flash solutions. The HGST has the SAS advantage what allows multipath io and HA solutions.

    Despite that, there is now a much faster solution with Intel Optane. Up ftom the 800P this will outperform any Flash solution. Only the powerloss aspect is different as Intel guatrantees this only for the top model 4800X. The lack of Dram caches makes the risk maybe tolerable with the smaller Optane.

    See my performance sequence where your HGST may be a comparable with the best Sata SSD Flash.
    http://openzfs.hfg-gmuend.de/napp-it_manuals/optane_slog_pool_performane.pdf
     
    #6
    Last edited: Aug 1, 2018
    nle and epicurean like this.
  7. epicurean

    epicurean Member

    Joined:
    Sep 29, 2014
    Messages:
    543
    Likes Received:
    20
    Thank you everyone. I wish I had the cash for an Optane now but I guess these HGST are the best I can do for now
     
    #7
  8. weust

    weust Member

    Joined:
    Aug 15, 2014
    Messages:
    287
    Likes Received:
    23
    Sorry to crash in on the topic, but with the Intel DC 3700, is that specifically 3700 or the series?
    Meaning, just 3700 or is 3710 fine too?
     
    #8
  9. i386

    i386 Well-Known Member

    Joined:
    Mar 18, 2016
    Messages:
    1,659
    Likes Received:
    400
    s3710 is a newer model, with sligthly better performance.
     
    #9
    weust likes this.
  10. pricklypunter

    pricklypunter Well-Known Member

    Joined:
    Nov 10, 2015
    Messages:
    1,514
    Likes Received:
    431
    For a good SLOG disk, look for one that has really low latency, certainly much lower than any storage pool of spinners that you plan on using it with, is designed/ optimised for high write endurance, and has power loss protection that's guaranteed to protect any inflight data. Watch that last one, read the small print carefully, lots of them claim some form of PLP, but not guaranteed to protect inflight data :)
     
    #10
  11. rune-san

    rune-san Member

    Joined:
    Feb 7, 2014
    Messages:
    78
    Likes Received:
    15
    For those without Optane money, or older platforms, or lacking PCI-e slots for Add In Cards, the HGST 100GB SAS drives are still *great* options for SLOG drives. I still use them myself in my Storage node. As gea mentioned, the real thing to ask yourself is if you really need Sync writes. Turn off sync writes where you can to get the most out of your platform. VMs, Databases, and other situations where files are opened on the node and kept open for work are good candidates for Sync Writes. Storage of files in general however is not. You of course can use Sync writes, but it's just slowing your performance.
     
    #11
    nle likes this.
  12. nle

    nle Member

    Joined:
    Oct 24, 2012
    Messages:
    179
    Likes Received:
    6
    What happens if you loose the slog? Will the pool still function?

    Also what about cache/z2arc?
     
    #12
  13. BackupProphet

    BackupProphet Well-Known Member

    Joined:
    Jul 2, 2014
    Messages:
    784
    Likes Received:
    278
    Yeah the pool will function just fine, just become a lot slower if you have many sync writes.
    If you lose the slog after an unclean shutdown you will lose everything that was written to the slog. Your pool will still be consistent and work fine, but some applications may fail to work properly because of data corruption.
     
    #13
    nle likes this.
  14. seizedengine

    seizedengine New Member

    Joined:
    Aug 22, 2016
    Messages:
    18
    Likes Received:
    0
    Note that on the S3700/S3710, the larger capacity ones are much faster. In my testing the 100GB HGST drives mentioned in the OP were faster than 200GB S3700s as SLOGS. I think Geas own testing showed this too.
     
    #14
  15. i386

    i386 Well-Known Member

    Joined:
    Mar 18, 2016
    Messages:
    1,659
    Likes Received:
    400
    intel s3700/3710 and hgst husmm use the same controler (developed by an intel & hgst joint venture), with the intel ssds using sata and hgst sas as the interface.
     
    #15
Similar Threads: SLOG drives
Forum Title Date
Solaris, Nexenta, OpenIndiana, and napp-it WD Ultrastar DC SS 530 SAS SSD for high performance HA storage/Slog Jan 10, 2019
Solaris, Nexenta, OpenIndiana, and napp-it ZFS - sync=disable - fragmentation without SLOG Jan 7, 2019
Solaris, Nexenta, OpenIndiana, and napp-it Intel Optane (32G/800P/900P) for ZFS pools and Slog on Solaris and OmniOS May 14, 2018
Solaris, Nexenta, OpenIndiana, and napp-it Partition same NVME for both SLOG and L2ARC May 6, 2018
Solaris, Nexenta, OpenIndiana, and napp-it ZFS performance vs RAM, AiO vs barebone, HD vs SSD/NVMe, ZeusRAM Slog vs NVMe/Optane Dec 6, 2017

Share This Page