In Late February, VMWare announced the release of ESX Server 3i, the first release of their embedded hypervisor platform for OEMs. Their initial list of partners who agreed to offer 3i hypervisors embedded in their servers is impressive -- Dell, Fujitsu Siemens, HP and IBM.
In Late February, VMWare announced the release of ESX Server 3i, the first release of their embedded hypervisor platform for OEMs. Their initial list of partners who agreed to offer 3i hypervisors embedded in their servers is impressive -- Dell, Fujitsu Siemens, HP and IBM. Hewlett-Packard will begin shipment of servers using 3i within the next week, in ten distinct models, and it is expected that similar announcements from the other Tier 1 system vendors will be made within the next few weeks.
There are certainly a lot of reasons why you’d want hypervisors to be embedded in firmware -- obviously tying it closer to the hardware platform gets you around a number of integration and performance issues, and it makes sense for it to be more aligned with the BIOS than the OS itself. It’s certainly evident to me that this is the wave of the future, where every server will have an onboard hypervisor and virtualization and virtual infrastructure becomes the de jure methodology for all forms of distributed computing. I can also see this extended even further to PCs, laptops and embedded devices, where every OS will be virtualized. From a development, support and deployment standpoint, having the OS abstracted completely from the hardware makes an awful lot of sense.
But do we really, really want a single vendor using a completely closed system calling the shots of what can and can’t run on our virtualized infrastructure? As good as the ESX 3i product is – and it is an excellent product – it is also a totally closed system and unless you are a Tier 1 hardware vendor, it is a black box. Yes, VMWare has the Open VMWare Tools project so that the Guest OS interfaces themselves to the hypervisor are open, but the inner workings of ESX are closed to all but a scant few. Gee this is starting to sound like, uh… mainframes and midrange computing.
Okay, perhaps I am starting to sound like I’m biting the hand that feeds me. Sometimes, proprietary hypervisor architectures make a lot of sense. In the mainframe and midrange UNIX computing world, such as on the IBM zSeries and pSeries it works, because you have a very controlled, highly optimized platform that is completely run by a single vendor. That extremely tight integration is what yields stability and performance, and certainly VMWare must be thinking along the same lines.
However, in the case of x86 computing where the environment is largely heterogeneous, it doesn’t make sense to completely lock down the hypervisor platform. What we really need is an open x86 server architecture that allows the customer to swap in and swap out any embedded hypervisor they want. Right now, nobody other than VMWare has something that is marketable as an embedded hypervisor, so maybe this entire argument sounds a bit premature. But presumably, once the Xen-based vendors as well as Red Hat, Novell and Microsoft have a solution to allow their hypervisors to be embedded into the firmware, this is going to make a lot more sense.
Am I alone in thinking that ESX 3i should be a choice, not a default? Talk back and let me know.