VMware, Citrix Xen, and Microsoft Hyper-V. Managing a Mix and Match Virtualization Platform Environment
One of the most interesting pieces of feedback that I got from my friends in enterprise IT shops that attended the recent VMworld conference in Las Vegas was that most of them expect to have more than one virtualization platform in the very near future. For most of them very near future means in the next year. The most common reasons given for this were:
- Despite everyone’s significant commitment to VMware, and VMware’s current acknowledged market leadership, many people felt underwhelmed by the keynotes and product roadmaps. Basically the feedback was that it was good, but it was not good enough for people to remain comfortable betting the operation of their entire data center on VMware. Bringing in another vendor is viewed as an insurance policy against VMware stumbling in some major aspect of its product and business strategy.
- Lot’s of people believe that given VMware’s lead in functionality that the higher prices for VMware are justified. However, they also want negotiating leverage with VMware, and view brining in credible alternatives as the best way to get that.
- The "next" things to get virtualized are in some cases different enough so that people are willing to look at alternatives that are the best fit for these next projects.
I had a detailed conversation with a friend who is the lead VMware architect at a huge international enterprise who has the following situation. They are 100% VMware today for the servers that they have virtualized in their data center. They have offices all over the world that have servers in them, none of which are virtualized today. They have a large farm of Citrix Presentation Servers (XenApp), none of which are virtualized today. Finally they are embarking upon a VDI project to slowly replace the XenApp environment with virtualized desktops. At this point the most likely virtualization platform decisions for these next projects are:
- When they upgrade the servers in the remote offices to Windows Server 2008, the three or four servers in each office will be collapsed into one physical server running Microsoft Hyper-V. Those servers are low stress servers like file, print and domain controllers for relatively small numbers of people, and there is no economic case for doing anything else. This means going from one platform to two.
- Since they are a large Citrix shop today, they are comfortable with Citrix products assuming those products perform. They have become convinced that XenServer is a better platform for virtualizing XenApp Servers than VMware. The basic feeling is that if Citrix gets one and only one thing right with XenServer it would be as a platform for XenApp. This means going from two platforms to three.
- Finally the VDI project. Right now the plan is to put this on the VMware back end, but use Citrix XenDesktop and the Citrix communications and security infrastructure (Citrix Access Gateway) to manage the VDI environment. Again the feeling is that VDI is a logical extension of the existing Citrix Presentation Server business and that Citrix has an enormous leg up in terms of in market experience and seasoned technology. This bumps the number of platforms up to four, with the last one actually being a mixture of VMware and Citrix.
Upon hearing this story from this friend and many other stories like it, my immediate reaction was "How the heck are you going to manage that"? The management strategies from the virtualization platform vendors (VMware, Microsoft, and Citrix) are simply not aligned with this use case. All of them are for the most part trying to use their management tools to differentiate their virtualization platforms. This is what the platform vendors have to do, since the hypervisor has gotten commoditized, so the management tools are the principle thing in the product portfolio that the platform vendors can charge for.
Of course my friend turned the question around to me. What would be my recommendation? The answer is what the rest of this article is about. I strongly believe that as the virtualization market grows and matures, the correct choice for management tools will for the most part be from vendors other than the platform vendors themselves. If you think about the number of management tools that the average enterprise owns (include all management and monitoring tools when you count), it is easy to see that history is in my favor on this one. How many network tools do you own that do not come from Cisco? How many performance and availability tools do you own that do not come from your hardware vendor, OS vendor, middleware vendor, database vendor, etc.
If you accept my thesis that you will need tools from vendors other than from virtualization platform vendors, the next question will be whether or not you need new tools from new vendors, or whether your existing management vendors will get there in time. As is almost always the case when a disruptive change happens like virtualization, the incumbent vendors are almost always to slow to be there in time. Do you really think that IBM/Tivoli, CA, BMC, and HP/Mercury are going to get there in time to help you manage the performance and availability of your new multi-platform virtualization environment? I don’t.
There is one more wrinkle that dramatically shifts this debate in favor of new vendors with new approaches. Not only is virtualization a new market opportunity, but when it comes to applications performance management and end user experience management, it is a disruptive change that breaks many of the incumbent approaches. I wrote a comprehensive analysis of this issue on my web site. The article is here if you would like to read it.
If you want the Reader’s Digest version of why incumbent approaches get broken here it is. Virtualization messes with the Guest OS’s perception of the clock. This shifts all time based metrics collected with a Guest by a random amount driven by the frequency and degree to which that Guest has been scheduled out by the hypervisor. If you would like to read VMware’s white paper on the subject it is here.
The time shifting problem breaks all incumbent products that use agents in Guests to collect things like CPU utilization, disk I/O rates, network I/O rates, context switches per second, page faults per second, and transaction response times from within the Guest. So a totally new approach needs to be taken to the problem of applications performance management in virtualized environments. This combined with the multi-vendor environment that many enterprises are going to, creates the need for an entirely new set of management tools.
So, which tools should you look at? I am not going to try to give a comprehensive list here, merely some examples of some tools that meet certain needs:
- If you are concerned about how your SAN and your virtualized servers are interacting in ways that can impact applications performance and end user experience (you should be), then Akorri has a great product for this need. Akorri deeply instruments the SAN and the key applications that use the SAN down to the level of the spindle and the LUN’s. It works across VMware, Citrix and Microsoft.
- If you want to understand how your applications are distributed across your virtualized environment, BlueStripe Software automatically discovers your applications topologies (and updates this as things get moved), and calculates application response time end to end and by layer across your applications. Just imagine a situation where your XenApp servers are on XenServer, and the rest of the application system is on VMware and there is a performance problem. Citrix EdgeSight is based upon the wrong architecture (agents in Guests), and does not look at response time across application layers. AppSpeed (formerly B-hive) will do this for certain types of applications, but is not shipping yet, and is not certain to support all of your platforms.
- If you are doing virtualized desktops, you have a unique set of issues. This starts with knowing who the users actually are, knowing what their end point device is, and knowing what their virtualized guest is. vmSight has a unique and patented technology, Connector ID that seamlessly provides this information while also giving you per user and per application response time information across all layers of all applications. vmSight even allows you to set policies that govern which users use which devices to access which back end resources. vmSight supports VMware and Citrix, and will support Hyper-V by year end.
- When it comes to management outside of applications performance and end user experience, the most pressing issue is how to stay on top of such a complex and rapidly changing environment from an administration perspective. In many organizations there are only one or two people who are allowed to create new VM’s, and no one has ever gone through all of the ones that have been created and killed the ones that are no longer needed. So, as virtualization goes to the next level in enterprises, a whole new way of managing it is needed. This new way will need to be based off of policy driven self-service provisioning and de-provisioning of VM’s. DynamicOps has built the product to meet this need and supports all three platforms (and combinations like XenDesktop/VMware) today.
In summary, if you know or think that you are going in any of the cross-platform directions outlined in this article, you need a cross-platform management strategy. The vendors that I mentioned above are a great place to start looking.
Application Performance Management Experts