Revelations of Windows Vista

I've often wondered how many people it takes to write a Windows kernel, and at last Microsoft seems to have provided the answer, which is a surprisingly small number.

My source is Rob Short, corporate vice-president in charge of the team designing the forthcoming Windows Vista kernel. In a recent interview Short said his kernel architecture team had six full-time members, and about 30 other part-time contributors. I'd be amazed if there were not more people working on the Linux kernel than there are in the Vista team.

Anyhow, Short was frank about the shortcomings of the Windows environment. For example, he conceded that NT is not a classic real-time operating system, and talked about the clash between engineering goals - his team is looking at the NT architecture over the next five or 10 years - and those of the marketing folk. He sees the Vista release as the first release of many, while marketing staff see the next release as the ultimate goal.

Short also talked about non-uniform memory access (Numa) support in Vista. At first glance this might seem like rather a high-end feature for a desktop operating system, but don't forget AMD's Opteron chips have a built-in memory controller. This means multiprocessor systems built with those chips, including dual-core versions, use Numa, so it's important for Microsoft to support it with a desktop operating system.

Given the performance bottleneck caused by using a shared memory bus for all processors, it's also possible that Intel could switch some of its chips to Numa at some point in the future. So it's good that Microsoft is making Numa support pervasive throughout its server and desktop lines.

Short also mentioned a new driver model that runs some device drivers in " user mode". As you know, today's software runs in one of several privilege levels within a processor. The kernel runs with the most control over the hardware.

Application software and some device drivers operate in user mode, which does not have the same level of control over the hardware. This means, for example, that a graphics driver running in user mode could not crash the entire system if it failed. However, it seems unlikely that Microsoft will add this capability to Windows Server software.

Short specifically said the approach would not be suitable for a server's network card device driver, for instance. As someone who remembers a time when NT was more stable than it is today, partly because at the time it did not use kernel mode network drivers, I think this is a pity.

In what seems to underscore the importance of Red Hat's Stateless Linux project, Short said the toughest problem for the architecture group is to improve the separation of state throughout the Wind- ows system. This state information includes the content of users' files and things like printer settings.

Short acknowledged that this information is not well structured in Windows and that this needs to change to allow Microsoft to really control the system in the ways it wants. It seems the differences in approach between Windows and Linux are not always as great as some would imagine.

Have your say, here: