8 Responses

  1. Matt
    Matt at |

    While I have never used Hyper-V, my company might one day simply because of the price. I do know of many companies that have switched to the V, and now most of them want to go back but they can’t. I agree with you that you don’t want all those VMs migrating at once. Not sure if the bug still exists, but there used to be one in VMware that would allow for more than 8 VMs to migrate at once which would saturate a network link. This is one reason we, we keep our VM traffic separate from other traffic.

    Reply
  2. Jim Millard
    Jim Millard at |

    Previous versions of vCenter had an undocumented (and therefore unsupported) config node that could be added to the vpxd.cfg file that would allow you to adjust the “max weight” a host could handle for simultaneous migrations (see http://www.boche.net/blog/index.php/2009/01/05/guest-blog-entry-vmotion-performance/ for the original guest post by Simon Long). Doing so permitted you to increase the number of simultaneous vMotions–at your own risk–at least in vSphere 4.
    If this node is still supported in v5, you have the opportunity to “diddle” with your environment and discover the sweet spot for migrations, especially when you also have the ability in v5 to use multi-NIC migrations, not to mention CNAs that report <10Gbps to ESXi while they actually provide 10Gbps performance. You also might find that your environment can actually complete mass-migrations faster if you LOWER the number of simultaneous migrations, especially if you're trying to evacuate more than one host at a time.

    Reply
  3. Hyper-V 3 Offers Unlimited Live Migrations — Kind of | VCritical

    […] with the upcoming release of Windows Server 2012.  Soon, Hyper-V administrators will be able to migrate an unlimited number of virtual machines at once.  How’s that for amazing: going from one concurrent live migration in the currently shipping […]

  4. Matt Vogt (@mattvogt)
    Matt Vogt (@mattvogt) at |

    This is how LA traffic operates. Sure, go ahead and put 2million cars on the road all at one time. What? You wanted to get somewhere on time? It’s just not a good idea.

    I can’t imagine for latency sensitive workloads (OLTP, etc.) that this wouldn’t bring it to it’s knees.

    Reply
  5. Josh Atwell (@Josh_Atwell)
    Josh Atwell (@Josh_Atwell) at |

    Great post Chris. I see this as another example of blind leading the blind where marketing/sales sees a “limitation” and doesn’t care WHY it is a limitation, only that their product does not have that limit. Sadly we see this same type of thing happen in the mobile phone platforms. The competition never validates WHY the limitation is there on the competitor, just that its there. In this instance I’d be very curious to see benchmarks and uptime as you start to exceed the VMware limits on a Hyper-V system. My money says those migrations will take a VERY long time and leave those VMs susceptible to issues as you pointed out.

    Our environment performs nearly 10,000 successful vmotions each week. That’s a ton of migrations and are a result of a variety of reasons. DRS and Maintenance being the primary of course. I’ve witnessed firsthand some “accessibility” impact when you’re pushing these limits. The app servers do not disconnect but you can definitely see the effects of link saturation. I could only imagine what the effects would be if we pushed those boundaries to Unlimited.

    Jim in the comments touched on another important element. Keep in mind that most CNA are 10GBPS but do not always provide that for vMotions. vSphere does not verify that it has 10gbps available to it when upping it’s vmotion limits. It simply recognize that a 10G link is there and sets the max accordingly. If you’re running a pair of CNAs on a blade server (UCS of course! 🙂 ) then you’re sharing that bandwidth with other services. A good admin will of course design this configuration so that vMotion does not effect other communication. Doing so means that while you could theoretically hit 8 simultaneous vMotions on a 10G link, you probably are limited to fewer than that. If you don’t have those limits in place I’d highly recommend you benchmark that the next time you hit up a maintenance mode. Just imagine if you kicked off 20 VM migrations at the same time with Hyper-V with only 4GB available on your 10G link. You’d be crawling like a baby and getting a few phone calls I’d imagine.

    My money says that they’ll pull a Sprint and drop that unlimited to 6 within 90 days of release of Hyper-V 3, but what do I know. I’m an iPhone on AT&T guy. I’m used to limits 😉 (note, still on the old iphone unlimited plan and have NEVER exceeded 2GB/month)

    Thanks for the post Chris! Excellent as always.

    Reply
  6. Richard Priest
    Richard Priest at |

    Hence why at TechEd North America the following statement was made: “Concurrent Migration: Limited Only By Hardware Resources”. Figure out what your hardware can cope with, and use that. The only person suggesting that it’s a good idea to try to migrate all your production workloads in one go on Hyper-V is Eric Gray. While I respect his knowledge of Hyper-V, he works for VMware so he’s clearly not he place to go for good advice on this.

    http://channel9.msdn.com/Events/TechEd/NorthAmerica/2012/VIR309

    Reply
  7. vSphere 5 versus Windows Server 2012 Hyper-V: live migrations | UP2V

    […] See the opinions of other bloggers like Chris Wahl in his posting titled From The Bad Ideas Department: Unlimited VM Migrations […]

  8. Frank Gia
    Frank Gia at |

    +10000 Chris. How comfortable would you be supporting ‘unlimited’ for a customer? 🙂 Glad the NS is starting to be seen for what it is.

    Reply

Share your point of view!