none
How to make a Hyper-V 2012 cluster perform RRS feed

  • Question

  • Hello,

    I am in the final stages of selecting components for Hyper-V 2012 cluster and SAN, I am interested in people's opinions on what the best tweaks for improved performance.

    SAN will be 8G FC (eg 3PAR 7400 / Compellent SC8000 / NetApp VNX 5500) with 30-40 Tier-1 disks, 20-30 Tier-2 disks, possible some SSD.

    Servers will be 2 CPU (eg E5-2680), 256GB Memory (16x16 1600Mhz), 3 x 146 15K in RAID-1 + spare, 2 x single port 8Gb HBA, 4 x 10GB NIC, 4 x 1GB NIC

    Introducing 10Gb switching for east-west traffic that will most likely be connected to 6509-E core using multiple 1Gbps.

    Where would you spend money if you wanted to increase performance ?

    SSD for the SAN ?

    10Gbps from rack switching to 6509 core ?

    Upgrade CPU to E5-2690 ?

    Fusion IO ?

    One vendor suggested having SSD for the Hyper-V OS drives.

    I do realise I have provided little details on the workload, assume typical small business, 1000 users, 60 guests, file, print, exchange, sql, sharepoint. Backup is on seperate hardware using DAS.

    Thanks

    Wednesday, May 22, 2013 6:24 AM

Answers

  • "perform" is a very qualitative thing.

    You have Network and disk and CPU and RAM.  They all have an impact.

    CPU and RAM is pretty straightforward.  Faster = more 'performance'

    Network is also straightforward, don't constrain it - this means that if you use PCI cards, make sure that your PCI bus is fast enough, don't just toss 10GB out, be sure it is all there through the stack.

    Disk - this is the hard to figure one.  I see lots of folks say "toss SSDs at it".  But, even MSFT in their Storage Story is getting smart here.  It is all about IOPS.  Which is about spindles, speed, and heads.  Determine the IOPS of your workload.  Do the math, and determine how to configure your storage to support that load.

    Any other features such as CSV caching and the like simply end up as bonuses (in some scenarios).

    Jose Barreto tells this story really well now.


    Brian Ehlert
    http://ITProctology.blogspot.com
    Learn. Apply. Repeat.
    Disclaimer: Attempting change is of your own free will.

    Wednesday, May 22, 2013 3:41 PM
    Moderator
  • I would put your money in to the 10GBPs network as you will have all new options going forward for potential SMB storage.  The secondary spot would be the SAN, which can also be used as an SMB target if you put a couple of Windows 2012 servers in front of it in a cluster.  I use 3PAR v400.  The 7400 is very good and has been licensing than the v400 as it grows. Adding a small SSD tier would give you some good performance, but with the architecture of the 3PAR you will do well with enough SAS and SATA disks in an AO group.  

    The servers should have enough resources for most all workloads.  I don't see any upgrades needed there.  Does that leave you any remaining DIMM slots for future expansion?

    As far as the Fusion IO I don't think it is necessary for most server workloads.  If you were going to have a large VDI deployment then I would say yes as it would absorb the boot IO storms that can happen during certain times of the day.  For servers that usually does not happen since you are not regularly booting all servers at once.  The Small SSD tier on the storage will help for workloads that are regularly heavy IO hitters. 

    Rob McShinsky (VirtuallyAware.com)


    VirtuallyAware - Experiences in a Virtual World (Microsoft MVP - Virtual Machine)

    Wednesday, May 22, 2013 8:00 AM

All replies

  • I would put your money in to the 10GBPs network as you will have all new options going forward for potential SMB storage.  The secondary spot would be the SAN, which can also be used as an SMB target if you put a couple of Windows 2012 servers in front of it in a cluster.  I use 3PAR v400.  The 7400 is very good and has been licensing than the v400 as it grows. Adding a small SSD tier would give you some good performance, but with the architecture of the 3PAR you will do well with enough SAS and SATA disks in an AO group.  

    The servers should have enough resources for most all workloads.  I don't see any upgrades needed there.  Does that leave you any remaining DIMM slots for future expansion?

    As far as the Fusion IO I don't think it is necessary for most server workloads.  If you were going to have a large VDI deployment then I would say yes as it would absorb the boot IO storms that can happen during certain times of the day.  For servers that usually does not happen since you are not regularly booting all servers at once.  The Small SSD tier on the storage will help for workloads that are regularly heavy IO hitters. 

    Rob McShinsky (VirtuallyAware.com)


    VirtuallyAware - Experiences in a Virtual World (Microsoft MVP - Virtual Machine)

    Wednesday, May 22, 2013 8:00 AM
  • "perform" is a very qualitative thing.

    You have Network and disk and CPU and RAM.  They all have an impact.

    CPU and RAM is pretty straightforward.  Faster = more 'performance'

    Network is also straightforward, don't constrain it - this means that if you use PCI cards, make sure that your PCI bus is fast enough, don't just toss 10GB out, be sure it is all there through the stack.

    Disk - this is the hard to figure one.  I see lots of folks say "toss SSDs at it".  But, even MSFT in their Storage Story is getting smart here.  It is all about IOPS.  Which is about spindles, speed, and heads.  Determine the IOPS of your workload.  Do the math, and determine how to configure your storage to support that load.

    Any other features such as CSV caching and the like simply end up as bonuses (in some scenarios).

    Jose Barreto tells this story really well now.


    Brian Ehlert
    http://ITProctology.blogspot.com
    Learn. Apply. Repeat.
    Disclaimer: Attempting change is of your own free will.

    Wednesday, May 22, 2013 3:41 PM
    Moderator
  • Hi Mick377,

    I came across your post and I saw you mentioned VNX 5500 is amongst your considerations.

    I am working for EMC and if you would like to get in touch with EMC directly to discuss how we can address your requirements I would be happy to help you with that.

    My twitter handle is @AoifeEMCSquared if you would like to PM me, I'd be happy to help.

    Thanks,

    Cloud1985.

    Monday, May 27, 2013 11:09 AM
  • Brian and Rob - thank you for your responses.

    CSV caching does sound good.

    A podcast I recently listended to said it reduced VDI boot time by 7x !

    Tuesday, May 28, 2013 3:10 AM
  • Do not know if you've already made your decision and you have acquired the Storage.If you did not,this time to ask for proof of it.I tell my story in brief.

    2 months ago M1000e Blade acquireone M620 with 16 sheetswith 2 Xeon 2620 processors, 64GB of RAM and two 146GB SAS disks that are configured in RAID 1.Within the same solution acquired two Compellent Storage SC8000, the primary has Cache 128GB, 7200GB SSD, 41 600GB SAS disks 15KRPM. The secondary has 64GB of disk cache and Fata 24 7.2K RPM 2TB. The result, in both IOPS is very good, as the first reference to IOMETER gave about 40,000iops, but foryou to consider,NOT ALL ARE IOPS in the world of storage,also you need to own a good throughput.

    Once all set up and configured, we begin operational testing. Install 5 Blade's with Windows 2012 DataCenter and assemble a cluster for Hyper-V 3.0, when a saved Run a virtual machine with 32GB of RAM allocated, it took me more than 5 minutes, but the problems started there recently. Then make a snapshot of virtual disks which added 1.46 TB of physical space occupied, the result, more than 10 minutes making the virtual snapshot and was not performing any work, so there was no changes to your files.

    Of course, immediately, Storage probe the secondary and to my surprise, the metrics were almost traced, but considering the big difference that the latter had discs and considering that possessed half of the primary cache. Immediately raise a claim and here began copying thousands of tests with Dell staff, threw the blame on Windows and wanted to try with VMWare, which allowed them to do and the results were the same. The file copy test using copy and paste in Windows, they could not overcome the 200MBytes / s in MPIO and 300MBytes / s in single path, it is curious that worked best with only 1 path. The case escalated into support levels and after 6 WEEKS!, I said the performance was as expected, that Compellent was designed for multi-threaded and to preserve the bandwidth per connection, that after 6 BONE WEEKS tell me something I already knew, but I did lose a huge amount of time with evidence that they would not get anywhere.

    I forgot, the solution incorporated in the drawer Blade, 2 MXL Force10 Switch's 10GB for LAN, 2 Switch's MXL Force10 for the SAN. The SAN did not share its bandwidth with no other traffic other than the the Storage. The M620 Blade's were prepared with dual port Broadcom 57810 10GB PCI-E 8x sockets and the Storage controllers were connected via dual port NIC's Chelsio 10GB all about iSCSI.
    In the process of testing and analysis, specialists placed Compellent Copilot, Networking and Software and 6 WEEKS AFTER I got the response that prompted earlier.

    So before buying anything, first ask a test that can check and verify and check thoroughly NIC's in these processes, because Windows hides the actual speed generating a huge cache in RAM if you have one available, but at the end of the process , if you look at the NIC's of SAN, will see that these still work to actually end the process.

    The SAN NICs on servers, at no time could pass 2Gbps of continuous use and this only occurred in spikes, usually not exceeding the value of 1gbps.

    As data, were tested with Windows 2008 R2 SP1 and Windows 2003 SP2 with Intel X520 NIC, and even change in a leaf is a NIC Intel X520 by giving the same results. Even in one of the blade of the Blade's, is instead a SAN by NIC Intel X520 and the result was the same.

    I hope you forgive me spread my story, but I hope that if effected the purchase, do it in the best way possible.
    I want to clarify that the Storage, has an excellent performance in his job. As an example, I can say that the volume of 1.46TB, being doubled, only takes me about 25 minutes to generate your copy and Replay System's also very interesting.

    Excuse my English, I translated text with Google translator.
    Any questions you have about this that I share with you, you can send it to me to be the answer.

    Greetings.

    Saturday, July 20, 2013 11:34 PM