virtualboy blog

Matt McSpirit on Virtualisation, Management and Core Infrastructure

Combining Hyper-V R2, Intel Ethernet, Intel 5500’s and iSCSI Storage…

Combining Hyper-V R2, Intel Ethernet, Intel 5500’s and iSCSI Storage…

  • Comments 2
  • Likes

This is a combination of technology that I’m starting to see on a pretty regular basis.  People looking at virtualisation, are typically weighing up new hardware, and from a server perspective, it makes sense to go for the latest and greatest CPUs (from both Intel and AMD) because of their virtualisation optimisations.  When it comes to storage however, more and more customers are learning to appreciate the performance that iSCSI can bring, yet at a great price point when compared with Fibre implementations.  This isn’t obviously correct for all scenarios, but, typically the rule holds.  Stephen’s got a great example of this here.  Stephen’s also got a link to a test performed back in March last year where Microsoft and Intel, combining the technologies above, pushed over 1 million IOPS.

If you are thinking of going down this route, combining these technologies, then I’d advise you to have a look at recent information produced by Microsoft and Intel around this very combination.  The stuff has been floating around the web on various blogs, news sites and so on, for a good few while now, but today was the first time I’ve had chance to have a look in more detail, and the results are very encouraging indeed.

In terms of resource and information, firstly, let’s get the marketing document out of the way.  This brief highlights the key benefits when you go down the ‘combination’ yet doesn’t go into the nuts and bolts like some of the other resources do.  Among other things, it highlights technologies like Microsoft’s VMq and Intel’s VMDq and how this combination can accelerate network traffic, including iSCSI.  It’s a useful starting point though.

Going back to what I said earlier about the 1 million IOPS, if you look at Stephen’s post, one of the comments is around the fact that it’s 1 million IOPS with 512 byte blocks, which isn’t reflective of applications like Exchange and SQL, which have larger recommended block sizes, and as you increase the block size, the number of IOPS will decrease.  This isn’t to say the test is flawed, nor is it marketed to hit the headlines.  What it proves is, iSCSI is ready for the big time, provided it is specced accordingly, and when I say that, don’t instantly think “This must be the highest-end SAN, and the highest-end iSCSI HBA’s” to achieve anything near those levels – if you read Stephen’s post carefully, you’ll find that the test was run using the in-box Windows Server 2008 R2 iSCSI initiator, on an Intel 5500 (which ships as standard in most boxes now), and it ran over a regular 10GbE Intel Ethernet NIC.  Sure, 10GbE is expensive, compared with 1GbE, but if you need that level of performance, achieving it with iSCSI is a very valid alternative to the Fibre approach and gives you more options around passing iSCSI traffic into VMs than Fibre will.

To quote Stephen

Performance is not an issue for iSCSI – Sure, not every iSCSI stack can handle a million IOPS, but the protocol is not the problem. iSCSI can saturate a 10 GbE link and deliver all the IOPS you might need.

Performance is not an issue for software – Today’s CPUs are crazy fast, and optimized software like the Windows Server 2008 R2 TCP/IP and iSCSI stacks can match or exceed the performance of specialized offload hardware.

Storage vendors need to step up their game – Whose storage array can service a million iSCSI IOPS? Raise your hands, please! I can’t hear you! Hello? Anyone there?

Fibre Channel and FCoE don’t rule performance – I don’t know of a Fibre Channel SAN that can push this kind of throughput or IOPS through a single link. Even FCoE over the same 10 GbE cable can’t quite do it. If they are to stay relevant, they had better come up with a compelling advantage over iSCSI!

What you’ll also find from Stephen’s post, is that Hyper-V also performed admirably, with performance remaining similar to native throughout.

If you can, watch this webcast.  It’s available on demand, and presents to you, first hand, the results that Stephen has referenced above.



Comments
  • Well, Hitachi Data Systems can support that many IOP's!

  • You're spot on John, and what a great combination of technology it is! :-)

    As soon as i saw this message above, I knew it would be from you! ;-)

    Hope all is well!

    Matt

Your comment has been posted.   Close
Thank you, your comment requires moderation so it may take a while to appear.   Close
Leave a Comment