What Changes Have Been Made Under the Hood?

blinksumgreen said:

I've been looking around for information about Windows 7 in regards to what changes have been made to the core OS. Most of my searches end with an article about the user interface changes and it might briefly skim over the more technical aspects.

I'm looking for information more along the lines of what, if any, changes were made to the kernel, to the way the system does processor scheduling, or the way it handles virtual memory management or prevents deadlock. I'm mainly interested in how Windows 7 differs from XP. (7 and Vista are a lot alike, so I guess the real question is how did Vista differ from XP)

Obviously I don't expect you to do the research for me, unless you want to.

But if anyone could throw some links my way, or post the answers to some of these questions, I would really appreciate it.

Hello and welcome to SF :)

If you want to discuss anything based on the info logicearth suggested - please consider doing so here!

Yeah.. I found this video to really speak to what you are asking.. Certain changes, like the dispatcher lock riddle, can be very much felt by the end user..

Mark Russinovich: Inside Windows 7 | Going Deep | Channel 9

musiclover7 said:

Yeah.. I found this video to really speak to what you are asking.. Certain changes, like the dispatcher lock riddle, can be very much felt by the end user..

Mark Russinovich: Inside Windows 7 | Going Deep | Channel 9

The dispatcher lock improvements are generally only going to be "felt" in a measurable way on servers. Thread contention is not as big an issue on client systems - the processors spend the majority of their time idling.

There's a bunch of misinformation out there which indicates that environmental performance problems which manifest themselves as perceptible periodic "freezing" are somehow due to the dispatcher lock. Nothing of the sort. It's just another of those "jargon" things that went into circulation after Russinovich started talking about it.

Well.. from my perspective, something is different.. My system, when it ran Vista, used to freeze up for no reason at least a half dozen times a day.. Now in 7, this is a non-issue. Based on the way Russinovich describes it in this interview, it sounded like he was speaking directly to this issue. 'A single thread makes a call, and every other thread has to wait for that thread to finish...' I am not saying I am right, but I notice my system does not seem to stop to think anymore after loading Windows 7.

musiclover7 said:

Well.. from my perspective, something is different.. My system, when it ran Vista, used to freeze up for no reason at least a half dozen times a day.. Now in 7, this is a non-issue. Based on the way Russinovich describes it in this interview, it sounded like he was speaking directly to this issue. 'A single thread makes a call, and every other thread has to wait for that thread to finish...' I am not saying I am right, but I notice my system does not seem to stop to think anymore after loading Windows 7.

I don't doubt your observations but I'm confident that the problem you used to experience had nothing to do with the dispatcher lock or Vista's implementation thereof.

A user-perceptible delay is a lifetime as far as the OS dispatcher is concerned. It steps in every 15ms to decide which thread to run next - on every processor on a multiproc box - which is partly why even a moderately busy machine does "context switches" a few thousand times per second, every second.

A server with hundreds or even thousands of ready-to-run (RTR) threads can make the dispatcher's life hard because a "lock" was required in order to protect certain management structures from damage by concurrent access by multiple processors running (re-entrant) dispatcher code. That may last micro/milliseconds, but with all those RTR threads the effect was somewhat degraded performance.

It's not something you'd notice in an app or UI of a client machine though. That's the type of sync SNAFU commonly caused by problems in filter drivers (AV, firewall...) or disk queue issues.