The purpose of this article is to analyze where VMware is today and where they're going with regards to Citrix. However, we need to look back before we can look ahead. Therefore, let's begin with some brief history.
Part 1: The History of using VMware with Citrix
VMware started selling their products in 1999. By 2002 they were very mature, and the server versions of VMware were used to power the datacenters of some very high-profile companies. However, VMware was still a small player in the Citrix space. That all changed in November.
At Citrix iForum in November 2002, Citrix, IBM, and VMware put out a press release talking about the virtues of using VMware to host multiple virtual Citrix servers on a single physical server. Why would anyone want to do this? For years, Citrix consultants (myself included) have been recommending a "build out" strategy versus a "built up" strategy for large Citrix farms. Most large Citrix environments were built with rows and rows of blade servers or 1U "pizza box" servers.
During 2002, however, some people began to notice a strange phenomena when building large Citrix (or Terminal Server) servers. They noticed that they couldn't really get more than about 200 or 225 users on a server, regardless of the amount of memory or processors. For example, in a situation where two processors could support 100 users, four processors would support 200 users, but six processors would only support 210 users, and eight processors would only support 220 users.
The cause (which I believe was first uncovered by Compaq) turned out to be related to the fact that the Microsoft Windows kernel can only use 2GB of memory, regardless of how much physical memory is in a server. Therefore, certain elements of the kernel memory (specifically the “system page table entries”) would run out of space when somewhere around 200 sessions were running on a single server. All the memory in the world couldn't fix this problem.
This limitation was a big problem for hardware companies. For example, in late 2002, IBM did not yet have a solid blade strategy, but they did have warehouses full of x440 servers. These servers could scale up to 16 processors. However, they were completely useless in the Citrix and server-based computing markets since the “maximum” 200 or so users only required dual processor servers.
The solution was VMware. VMware's software allows multiple Windows servers (complete with their own BIOS settings, reboots, and network interfaces) to run together on the same physical server. Therefore, with these 16 processor x440 servers from IBM, customers could run perhaps 8 or 16 virtual Citrix servers on the same physical box, easily hosting 1000 or 2000 Citrix ICA sessions on a single IBM x440 server.
The iForum 2002 closing party at DisneyWorld was abuzz with talk of this solution. From a technical standpoint it worked like a charm. VMware deserves a ton of credit for their work with their solid products. Unfortunately, people were less enthusiastic about this solution from a practical standpoint. Did it make sense to buy a 16-way server and VMware software instead of 8 dual processor servers? Was the solution more manageable, or did it introduce a single point of failure? Would Microsoft even support a VMware solution?
This was complicated by the fact that while IBM was pushing their 16-processor servers, HP/Compaq was pushing their blades. (Dell was no where to be seen at that show.) As for manageability, HP was pushing an Altiris/ThinkDynamics solution that simplified the management of blade-based solutions so that they were on par with single server solutions (such as the x440 from IBM).
The general consensus leaving iForum in November 2002 was that the VMware-based solutions were great for Citrix if you already had a big commitment to VMware or if you had some particularly intensive applications, but in general you should stick to the “build out” strategy.
The next few months didn't produce any drastic news in this area, with a few small joint Citrix/VMware wins. However, things really heated up the second quarter of 2003.
April 2003 marked the release of Windows Server 2003. This was significant in the VMWare / Citrix space for one reason: Certain architectural changes in Windows 2003 lessened the impact of the 2GB kernel memory limitation. (For example, the number of system page table entries doubled from Windows 2000 to 2003, and 2003's registry was no longer stored in the kernel's memory area.) These architectural changes meant that vanilla Windows 2003 servers were no longer limited to the “200 or so” users like Windows 2000. Early testing by HP showed that a single Windows 2003 server could support 400 or 500 users in some situations—all without VMware.
Then, in May, Microsoft announced that it was purchasing the virtual computing software assets from a VMware competitor called Connectix. Connectix had been working on a virtual server product that they felt would rival VMware's GSX server. The effects of this announcement turned out to be the opposite of what most people thought. Instead of threatening VMware's position, Microsoft's acquisition of Connectix served to validate the virtual server approach. After all, how could Microsoft not support virtual servers if they themselves were doing it?
As 2003 comes to a close, Microsoft, VMware, and Citrix's virtual server strategies are maturing.Microsoft spent the second half of 2003 preparing their Virtual Server solution, currently scheduled for a 1Q 2004 release. VMware released some enterprise products that futher enhance uptime and availability. Check back Wednesday for Part 2 of this article (including an interview with VMware's VP of Marketing) that covers where the three companies are today and what we can expect in the future.