Yellow Bricks

by Duncan Epping

vmotion

vGPUs and vMotion, why the long stun times?

· · 2 Comments

Last week one of our engineers shared something which I found very interesting. I have been playing with Virtual Reality technology and NVIDIA vGPUs for 2 months now. One thing I noticed is that we (VMware) introduced support for vMotion in vSphere 6.7 and support for vMotion of multi vGPU VMs in vSphere 6.7 U3. In order to enable this, you need to set an advanced setting first. William Lam described this in his blog how to set this via Powershell or the UI. Now when you read the documentation there’s one thing that stands out, and that is the relatively high stun times for vGPU enabled VMs. Just as an example, here are a few potential stun times with various sized vGPU frame buffers:

  • 2GB – 16.5 seconds
  • 8GB – 61.3 seconds
  • 16GB – 100+ seconds (time out!)

This is all documented here for the various frame buffer sizes. Now there are a couple of things to know about this. First of all, the time mentioned was tested with 10GbE and the NVIDIA P40. This could be different for an RTX6000 or RTX8000 for instance. Secondly, they used a 10GbE NIC. If you use multi-NIC vMotion or for instance a 25GbE NIC than results may be different (times should be lower). But more importantly, the times mentioned assume the full frame buffer memory is consumed. If you have a 16GB frame buffer and only 2GB is consumed then, of course, the stun time would be lower than the above mentioned 100+ seconds.

Now, this doesn’t answer the question yet, why? Why on earth are these stun times this long? The vMotion process is described in this blog post by Niels in-depth, so I am not going to repeat it. It is also described in our Clustering Deep Dive book which you can download here for free. The key reason why with vMotion the “down time” (stun times) can be kept low is that vMotion uses a pre-copy process and tracks which memory pages are changed. In other words, when vMotion is initiated we copy memory pages to the destination host, and if a page has changed during that copy process we mark it as changed and copy it again. vMotion does this until the amount of memory that needs to be copied is extremely low and this would result in a seamless migration. Now here is the problem, it does this for VM memory. This isn’t possible for vGPUs unfortunately today.

Okay, so what does that mean? Well if you have a 16GB frame buffer and it is 100% consumed, the vMotion process will need to copy 16GB of frame buffer memory from the source to the destination host when the VM is stunned. Why when the VM is stunned? Well simply because that is the point in time where the frame buffer memory will not change! Hence the reason this could take a significant number of seconds unfortunately today. Definitely something to consider when planning on using vMotion on (multi) vGPU enabled VMs!

VMworld Reveals: vMotion innovations

· · 1 Comment

At VMworld, various cool new technologies were previewed. In this series of articles, I will write about some of those previewed technologies. Unfortunately, I can’t cover them all as there are simply too many. This article is about enhancements that will be introduced in the future to vMotion, the was session HBI1421BU. For those who want to see the session, you can find it here. This session was presented by Arunachalam Ramanathan and Sreekanth Setty. Please note that this is a summary of a session which is discussing a Technical Preview, this feature/product may never be released, and this preview does not represent a commitment of any kind, and this feature (or it’s functionality) is subject to change. Now let’s dive into it, what can you expect for vMotion in the future.

The session starts with a brief history of vMotion and how we are capable today to vMotion VMs with 128 vCPUs and 6 TB of memory. The expectation is though that vSphere in the future will support 768 vCPUs and 24 TB of memory. Crazy configuration if you ask me, that is a proper Monster VM.

[Read more…] about VMworld Reveals: vMotion innovations

How to disable DRS for a single host in the cluster

· ·

I saw a question today which was interesting, how do I disable DRS for a single host in the cluster? I thought about it, and you cannot do this within the UI, at least… there is no “disable DRS” option on a host level. You can enable/disable it on a cluster level but that is it. But there are of course ways to ensure a host is not considered by DRS:

  1. Place the host in maintenance mode
    This will result in the host not being used by DRS. However it also means the host won’t be used by HA and you cannot run any workloads on it.
  2. Create “VM/Host” affinity rules and exclude the host that needs to be DRS disabled. That way all current workloads will not run, or be considered to run, on that particular host. If you create “must” rules this is guaranteed, if you create “should” rules then at least HA can still use the host for restarts but unless there is severe memory pressure or you hit 100 CPU utilization it will not be used by DRS either.
  3. Disable the vMotion VMkernel interface
    This will result in not being able to vMotion any VMs to the host (and not from the host either). However, HA will still consider it for restarts and you can run workloads on the host, and the host will be considered for “initial placement” during a power-on of a VM.

I will file a feature request for a “disable drs” on a particular host option in the UI, I guess it could be useful for some in certain scenarios.

High latency VPLEX configuration and vMotion optimization

· ·

This week someone asked me about an advanced setting to optimize vMotion for VPLEX configurations. This person referred to the vSphere 5.5 Performance Best Practices paper and more explicitly the following section:

Add the VMX option (extension.converttonew = “FALSE”) to virtual machine’s .vmx files. This option optimizes the opening of virtual disks during virtual machine power-on and thereby reduces switch-over time during vMotion. While this option can also be used in other situations, it is particularly helpful on VPLEX Metro deployments.

I had personally never heard of this advanced setting and I did some searches both internally and externally and couldn’t find any references other than in the vSphere 5.5 Performance paper. Strange, as you could expect with a generic recommendation like the above that it would be mentioned at least in 1 or 2 other spots. I reached out to one of the vMotion engineers and after going back and forth I figured out what the setting is for and when it should be used.

During testing with VPLEX and VMs using dozens of VMDKs in a “high latency” situation it could take longer than expected before the switchover between hosts had happened. First of all, when I say “high latency” we are talking about close to the max tolerated for VPLEX which is around 10ms RTT. When “extension.converttonew” is used the amount of IO needed during the switchover is limited, and when each IO takes 10ms you can imagine that has a direct impact on the time it takes to switchover. Of course these enhancements where also tested in scenarios where there wasn’t high latency, or a low number of disks were used, and in those cases the benefits of the enhancements were negligible and the operation overhead of configuring this setting did not weigh up against the benefits.

So to be clear, this setting should only be used in scenarios where high latency and a high number of virtual disks results in a long switchover time during migrations of VMs between hosts in a vMSC/VPLEX configuration. I hope that helps.

What does support for vMotion with active/active (a)sync mean?

· ·

Having seen so many cool features being released over the last 10 years by VMware you sometimes wonder what more they can do. It is amazing to see what level of integration we’ve see between the different datacenter components. Many of you have seen the announcements around Long Distance vMotion support by now.

When I saw this slide something stood out to me instantly and that is this part:

  • Replication Support
    • Active/Active only
      • Synchronous
      • Asynchronous

What does this mean? Well first of all “active/active” refers to “stretched storage” aka vSphere Metro Storage Cluster. So when it comes to long distance vMotion some changes have been introduced for sync stretched storage. (** note that “active/active” storage is not required for long distance vMotion**)With stretched storage writes can come from both sides at any time to a volume and will be replicated synchronously. Some optimizations have been done to the vMotion process to avoid writes during switchover to avoid any delay during the process as a result of replication traffic.

For active/active asyncronous the story is a bit different. Here again we are talking about “stretched storage” but in this case the asynchronous flavour. One important aspect which was not mentioned in the deck is that async requires Virtual Volumes. Now, at the time of writing there is no vendor yet who has a VVol capable solution that offers active/active async. But more important, is this process any different than the sync process? Yes it is!

During the migration of a virtual machine which uses virtual volumes, with an “active/active async” configuration backing it, the array is informed that a migration of the virtual machine is taking place and is requested to switch from asynchronous replication to synchronous. This to ensure that the destination is in-sync with the source when the VM is switched over from side A to side B. Besides switching from async to sync when the migration has completed the array is informed that the migration has completed. This allows the array to switch the “bias” of the VM for instance, especially in a stretched environment this is important to ensure availability.

I can’t wait for the first vendor to announce support for this awesome feature!