none
Virtualization and configuration LUN

    Question

  • Can you tell me it is better to have one big LUN (~10 TB) or many smaller LUN (~2 TB) on virtualized environment, Windows Server 2012 and Windows Server 2012 R2 with Hyper-V?

    Most of servers have SQL, and four of these servers have four large database (size of one DB is around 700GB) and use transaction log and temp DB.

    We use this storage with configuration:

    EMC VNX 5300
    RAID 5
    SAS 10k RPM
    NTFS
    Around 20 VM (Hyper-V) with different role on Windows OS (IIS, SQL, etc.)
    We use Cluster disk, Cluster shared volume
    Tiering is used

    IBM V7000
    RAID 5
    SAS 15k RPM
    NTFS
    Around 20 VM (Hyper-V) with different role on Windows OS (IIS, SQL, etc.)
    We use Cluster disk, Cluster shared volume
    Tiering is used
    Thursday, April 06, 2017 8:23 AM

All replies

  • Hi Sir,

    IMHO , both of two methods are fine .

    Because , you can control I/O based on VM virtual disk level (in vm settings) .

    The only difference may be the disk space usage , if you put VHD to separate LUN and each VHD doesn't fill up the LUN .

     

    Best Regards,

    Elton


    Please remember to mark the replies as answers if they help.
    If you have feedback for TechNet Subscriber Support, contact tnmff@microsoft.com.

    Friday, April 07, 2017 8:47 AM
    Moderator
  • I agree with Elton.  Either works fine.  It comes down to an operational issue.  If you have some disaster that requires you to recover A LUN, would you rather recover a single 10TB LUN and be completely out of operation while you are recovering the LUN, or would you rather recover a single 2TB LUN and lose only those VMs that are dependent upon that one LUN that failed?  Backup windows can also come into play.  It may take longer to backup a single 10 TB LUN than to backup five 2TB LUNs.

    So you need to decide what works for your particular environment.  From the Hyper-V side, neither one is 'better' than the other.


    tim

    Friday, April 07, 2017 12:56 PM
  • Also just be wary about your RAID levels - RAID-5 is considered by many storage vendors (unless you are using it for Flash based storage arrays or LUNs) as a dead technology for spinning disks due to the risk that RAID-5 brings with URE's and failure rates.

    As a basis I always RAID-10 for production data on spinning disks, and RAID-6 for backup/archive data on spinning disks.


    Robert Milner | Website: http://www.remilner.co.uk | Twitter: @robm82

    Friday, April 07, 2017 3:00 PM
  • Besides the reasons mentioned above, also consider disk queues. Each smaller LUN will also have a separate disk queue, whereas a single LUN will only have 1. In my experience, spreading your VM workload across multiple LUNs will let the SAN run more efficiently and you most likely will get higher total IOPs from the SAN.
    Saturday, April 08, 2017 6:07 AM
  • Hi Elton_Ji,

    You say "you can control I/O based on VM virtual disk level (in vm settings)". Do you mean on Advanced Features QoS Management or?

    Saturday, April 08, 2017 6:07 PM
  • Hi Tim,

    What can happen with LUN that need to be recover? How can I lose my LUN? What is possible reason?

    Saturday, April 08, 2017 6:12 PM
  • Hi robm82,

    But I use RAID5 with striping and LUN. What do you think about that?

    Saturday, April 08, 2017 6:14 PM
  • Hi D.Pope,

    This applies also for one big LUN with RAID5 and striping and Hyper-V environment?

    Saturday, April 08, 2017 6:32 PM
  • Maty,

    Not sure I follow. But what I meant was that if you present multiple smaller LUNs to a server, you'll get a queue for each LUN. If you only present one large LUN, you only get one disk queue to go with it. That one queue then gets shared between all the VMs on the LUN.

    Saturday, April 08, 2017 10:32 PM
  • Hi Elton_Ji,

    You say "you can control I/O based on VM virtual disk level (in vm settings)". Do you mean on Advanced Features QoS Management or?

    Hi Maty,

    Yes .

    Best Regards,

    Elton


    Please remember to mark the replies as answers if they help.
    If you have feedback for TechNet Subscriber Support, contact tnmff@microsoft.com.

    Monday, April 10, 2017 12:42 AM
    Moderator
  • Hi Elton_Ji,

    This features is for Windows Server 2012 R2, but on some physical servers I have Windows Server 2012. How can I control I/O on that servers?

    Monday, April 10, 2017 7:38 AM
  • "What can happen with LUN that need to be recover?"

    A LUN can become corrupted.  You can have a multiple disk failure causing all data on the disk to be lost.


    tim

    Monday, April 10, 2017 1:26 PM
  • 2012 R2 introduced Storage Quality of Service, so you do not have that capability in 2012.

    It may sound like a flippant answer, but you should upgrade to 2012 R2.  There are many improvements and it better prepares you for moving to 2016 in the future.  In fact, if you are upgrading, you should even consider moving directly from 2012 to 2016 to get the most improvements to your environment.


    tim

    Monday, April 10, 2017 1:29 PM
  • RAID5 still works, but it becomes a bit more 'dangerous' when used with large disks.  Should a disk in the RAID set fail, the time required to rebuild the RAID set when the failed drive is replace can take a significant amount of time.  During this time it is possible to lose another disk due to failure, meaning everything on the RAID set is lost and it is time to go to your backup media to recover everything.

    It is also the slowest performing RAID environment.  This may or may not be an issue for you.  If you have a write-intensive environment, RAID5 is considered the worst way to present the storage.  All that said, I again say it still works.  It is just not a favored solution anymore.


    tim

    Monday, April 10, 2017 1:33 PM
  • Hi Tim,

    We want to use IBM v5020 Storwize with striping. Vendor told us that the best for this storage is to use for SSD and SAS disks RAID5 (MDisk) with big LUN. We have write-intensive environment and when we propose to configure RAID10 vendor told us that this recommendation is for old storage system. Vendor also told us:The bigger the pool, the better performance regardless of the type of discs.

    Can you comment this?

    Monday, April 10, 2017 1:44 PM
  • MatyM - 

    I would suggest that you look at the storage vendor's suggestions.  You mention Tiering is used which would refer to SSD or FlashMemory is used to first absorb your write and present the verification back to your SQL machines. The array then does the final writes down to the slower SAS drives.  The trick is that some vendors will assign a portion of what the whole array has to a single LUN.  

    As to the size of your LUN, that is really how much do you want to restore before you get your data back?  2TB worth or 10TB?  If you like having all your VMs in one logical unit on the array, then ok.  Personally I suggest breaking them apart.

    Monday, April 10, 2017 1:56 PM
  • While it is a true statement that the bigger the pool the better the performance, I have never before heard that RAID10 is for 'old storage systems'.  Of course, I may not have enough experience since I just recently retired after spending more than 40 years in this business. <grin>

    RAID10 has always been considered the fastest RAID set.  RAID5 has always been considered the slowest RAID set for write operations.  RAID5 requires a minimum of 4 IOs for each write operation.  RAID10 requires 1.  I don't know what sort of physics your vendor is talking about, but go look at any disk performance benchmark to see if you can find any where RAID5 write performance outperforms RAID10 write performance.

    Disk performance is a science.  We have no knowledge of what your requirements are or how the application performs IO.  I have dealt with a number of performance environments over the years, and any time a benchmark was deemed necessary to prove performance for a write-intensive application, RAID10 always outperformed RAID5.


    tim

    Monday, April 10, 2017 2:05 PM
  • Maty,

    Tim is absolutely right about the performance of RAID5 vs RAID10. You're original question was regarding LUN size, and I agree with Tim on that as well. Multiple smaller LUNs may take more administration, but allows much quicker recovery in case of an issue. Look at what your RTO and RPO are and make the decision for your situation.

    Also, assuming IBM 'Easy Tiering' actually manages incoming writes and the location of 'hot' data well; AND your write workload fits in their cache/fast tier, you will never notice the write performance penalties of RAID5 for the capacity tier.

    Monday, April 10, 2017 2:16 PM