Posted By: The Channel 9 Team | Apr 4th, 2005 @ 11:54 PM | 87,788 Views | 33 Comments
Here's the final part of the interview we did with Windows kernel architect Dave Probert.

In this part Dave discusses the differences between the NT architecture and other operating systems.

Dave talks about the history of operating systems at Microsoft (not to mention Posix, OS/2, Unix, and other OS's). Interesting stuff, first time we've had a kernel architect on Channel 9.

If you wanna understand how Windows works, you gotta watch these videos. Thanks Dave for taking some time out of your day to explain all of this.

Here's the entire interview's parts:

Part I
Part II
Part III
Part IV (this thread)

Tags: Kernel, OS
Media Downloads:
Rating:
1
0
I would definitely like to see more information about the Object Manager and how the subsystems work (e.g. how a process chooses a subsystem and what implications that has on the interfaces available to it). Also, maybe some discussion on the future of subsystems (ok, Win32 "won" -- does that mean the concept of multiple subsystems is somewhat obsolete and from now on we'll all build on top of Win32, or does it make sense to have, say, a .NET subsystem if .NET is really going to supercede Win32?).
This series is the best ever... I love it... Thanks C9!

mrichman
mrichman
Mark A. Richman
It's my understanding that one of the architectural goals of WinFX is to bypass Win32 API completely, and have .NET 2.x talk right to the kernel.
PerfectPhase
PerfectPhase
"This is not war, this is pest control!" - Dalek to Cyberman
More, More!

That was by far the best series of video I have ever seen on Channel9.  Please do more like this!

Stephen.
Charles
Charles
Welcome Change
PerfectPhase wrote:
More, More!

That was by far the best series of video I have ever seen on Channel9.  Please do more like this!

Stephen.


Glad you like this! Stay tuned, you'll see more of the kernel in the near future. Next up? IO Manager and Cache Manager. Then, dig into the Memory Manager and Object Manager. Let's not forget about File Systems, eh.

I love this stuff too which is why you'll be seeing more and more deep stuff around here.

C
CplCarrot
CplCarrot
Dust Puppy
Charles wrote:
Glad you like this! Stay tuned, you'll see more of the kernel in the near future. Next up? IO Manager and Cache Manager. Then, dig into the Memory Manager and Object Manager. Let's not forget about File Systems, eh.

I love this stuff too which is why you'll be seeing more and more deep stuff around here.

C


Ahmen to that but can you get them to include the future work as well as the history - I was just screaming about indego at the screen.

Charlie
Charles wrote:
PerfectPhase wrote: More, More!

That was by far the best series of video I have ever seen on Channel9.  Please do more like this!

Stephen.


Glad you like this! Stay tuned, you'll see more of the kernel in the near future. Next up? IO Manager and Cache Manager. Then, dig into the Memory Manager and Object Manager. Let's not forget about File Systems, eh.

I love this stuff too which is why you'll be seeing more and more deep stuff around here.

C


Any GUI stuff in the future? We got a brief overview
already...
It really amazes me that something as complex as NT can be demystified to this point that even an undergrad college student with a decent OS class can understand a lot of these intricacies! 

I'm really getting the sense though, that the whole argument in favor of garbage collection is really kinda overblown.  If ref counters are used at the OS level with a good deal of success, there should presumably be a way in .NET to turn off the garbage collector and use an XYZ ref counter system instead for real-time/networking kind of stuff...which is what seems problematic when everything goes the way of .NET (even real-time programming) in Longhorn.

There really ought to be a better way to schedule than using various garbage collection algorithms (Java, .NET, etc.) which take out memory after the fact at some semi-random time in the future...maybe it's an NP-hard problem?
PerfectPhase
PerfectPhase
"This is not war, this is pest control!" - Dalek to Cyberman
Pure reference counting systems have issues as well, such as if two objects take out a reference on each other then the reference count will never drop to zero, hence why you need reference tracking systems such as the garbage collector that can work out these circular dependancies.

The is a paper somewhere, by Chris Sells I think where he modifies the Rotor code to use a hybrid system of counting and tracking.  Ah, here it is http://www.sellsbrothers.com/spout/default.aspx?content=archive.htm#refCountRotor

Stephen
Microsoft Communities