WilliamLam.com

VMworld Barcelona #NotSupported Tips/Tricks for vSphere 5.5 Slides Posted

by // 3 Comments

I hope everyone got back home quickly and safely from VMworld Barcelona (unlike myself for those of you who follow me on Twitter, took a bit longer than I had hoped). For those of you who attended my #NotSupported Tips/Tricks for vSphere 5.5 at VMworld Barcelona earlier in the week, I have posted my Sliderocket presentation below as well as the recording thanks to the vBrown Bag team. If you could not attend the session or did not go to VMworld you can view all vBrown Bag Tech Talk recordings by checking out this link here. I highly recommend you check out the ENTIRE presentation as there was an exciting announcement that I made at the VERY END of the presentation.

Disclaimer: I think it should be pretty obvious, but things discussed in the presentation is not officially supported by VMware. Use at your own risk.

I would like to thank everyone that attended my session, really enjoyed the crowd and the questions/discussions afterwards. I know you could have been else where such as the Solution Exchange with a nice beverage, so thank you all for attending and hopefully everyone enjoyed it. I would also like to give a big shout out to vBrown Bag team Jon Harris, Kyle Murray, Damian Karlson and Gregg Robertson for putting together such an awesome event up for the VMware community and for their assistance on quickly getting the AV all setup for my presentation.

Here is the presentation:

Here is the video recording:

Here is a couple of pictures from the audience:

Quick Tip: New Hyper-V guestOS identifier in vSphere 5.5

by // 16 Comments

For those of you who are so inclined to run Hyper-V as a nested VM on ESXi 5.5 (not sure why anyone would want to do such a thing), you should be aware that there is a new guestOS identifier that you can configure your VM to for the most optimal configuration. The main reasons you would want to use this configuration is that by default Windows Enlightenment is enabled and this will prevent Hyper-V from running as it will detect it is running inside of a VM. This configuration will disable Windows Enlightenment to allow you to run Hyper-V.

I noticed a new guestOS identifier called "windowsHyperVGuest" while browsing through the vSphere 5.5 API Reference guide, but when I checked the vSphere Web Client, I did not see this guestOS type as an available option. Perhaps this was not a supported guestOS, after all Nested Virtualization is not officially supported by VMware. In any case, you can still configure your VM by leveraging the vSphere API.

Here is a quick vSphere SDK for Perl script called changeGuestOSID.pl which allows you to reconfigure a VM with a valid guestOS identifier from the vSphere API Reference guide. You can of course easily do this using PowerCLI as well as any other language that can speak to the vSphere API.

Once updated, you should now see it reflected in both the vSphere Web/C# Client:

Note: I did not do extensive testing other than basic installation of latest Hyper-V Server and I do not believe you need any additional settings. If you wish to run nested 64-bit VMs, then you will need to enable VHV.

Ravello: An interesting solution using Nested Virtualization

by // 6 Comments

As many of you know, I am a huge fan of VMware Nested Virtualization and I am always interested to learn how customers and partners are using this technology to help enable them to solve interesting problems. I recently met up with a startup company called Ravello who has a product that leverages Nested Virtualization in a very unique way.

Note: Ravello is not the only company using Nested Virtualization in interesting ways. Bromium, another startup in the security space, is also doing interesting things with Nested Virtualization.

Ravello is a SaaS solution that allows you to take an existing VMware or KVM virtual machine and without any modifications to that VM, run it on a variety of public cloud infrastructures including Amazon EC2, HP Cloud, Rackspace and even private clouds that are running on vCloud Director (support coming soon). Ravello is basically "normalizing" the VM by virtualizing it in their Cloud Application Hypervisor so that it can run on any cloud infrastructure.  From the diagram below, the unmodified VM is actually running inside of another VM which runs a flavor of Linux. This Linux VM loads up their HVX Hypervisor and is running on one of the public cloud infrastructures.

Similar to a regular hypervisor, HVX provides an abstraction, but instead of the underlying physical hardware it abstracts away the underlying cloud infrastructure. The HVX hypervisor provides the following three core capabilities:

  • Presents a set of virtual hardware that is compatible with VMware ESXi, KVM and XEN virtual machines
  • Virtual networking layer that is a secure L2 overlay on top of the cloud infrastructure L3 networking using a protocol similar to GRE but running over UDP
  • Cloud storage abstraction that provides storage to the VM through Ravello Image Store that can be back-ended by Amazon S3, CloudFiles or even block/NFS volumes

My first thought after hearing how Ravello works, is that this is pretty neat! Of course the next logical question that I am sure most of you are asking is how is the performance? We know that running one level of Nested Virtualization will incur some performance penalty and this will continue with additional levels of Nested Virtualization. Ravello is also not leveraging Hardware-Assisted Virtualization but Binary Translation (a technique developed by VMware) as that can not be guaranteed to be available on all cloud infrastructures. In addition to Binary Translation, they are also using various techniques such as caching and chaining translated code, fast shadow MMU, direct execution of user space code and few others to efficiently run in a nested environment.

I was told that performance was still pretty good and sometimes even out performing regular cloud infrastructures. There was no mention of specific applications or performance numbers, so I guess this is something customers will need to validate in their own environment. I am also interested to see what the overhead is by doing two-levels of Nested Virtualization and what impact that has to the guestOS and more importantly, the applications. To be fair, Ravello's current target audience is Dev/Test workloads, so performance may not be the most critical factor. They also provide two modes of deployment based on cost optimized or performance and if the latter is selected, overcommitment of resources or consolidation will not be used.

Overall, I thought Ravello's solution was pretty interesting and could benefit some customers looking to run their workloads in other public cloud infrastructures. I think performance is just one of the things customers will need to consider but also how do they go about managing and operating this new VM container and how tightly integrated is Ravello with the VMware platform or other hypervisors for that matter. Though the VM and the underlying applications does not need to change, what operational challenges does this introduce to administrators? 

Ravello also recently presented their HVX Cloud Application Hypervisor at a recent USENIX conference and you can find more details in their presentation called HVX Virtualizing Cloud along with their research paper which can be found here.

One thing that I did want to point out after watching the presentation is that one of the presenter mentioned that their HVX nested hypervisor runs more efficiently than any other hypervisor out there and that others would require things like Intel's VMSC Shadowing feature to be comparable. I can not speak for other hypervisors, but when running VMware Hypervisors on top of our ESXi Hypervisor, our hypervisor has already been optimized for VMREAD/VMEXITS and Intel's VMSC Shadowing feature would only benefit slightly. You can read more about those techniques in this blog article.

Ravello will be at VMworld US booth #425 and I will probably drop by for a demo to see their solution in action.