Strange hang/CPU utilization issue on dual-core XP SP2 system - Microsoft Windows

This is a discussion on Strange hang/CPU utilization issue on dual-core XP SP2 system - Microsoft Windows ; This is a strange one. I had been doing some file organizing, and this involved fairly common Explorer right click -> new -> new folder operations. Meantime as I participate in distributed computing there's two low-priority background tasks each chewing ...

+ Reply to Thread
Results 1 to 3 of 3

Thread: Strange hang/CPU utilization issue on dual-core XP SP2 system

  1. Strange hang/CPU utilization issue on dual-core XP SP2 system

    This is a strange one. I had been doing some file organizing, and this
    involved fairly common Explorer right click -> new -> new folder
    operations. Meantime as I participate in distributed computing there's
    two low-priority background tasks each chewing 100% of one core.

    Then I right click -> new and Explorer simply hangs. First thing I
    notice is the submenu failed to appear on schedule. After a moment I go
    to the task manager only clicking the task manager (already open) in the
    task bar doesn't work -- and the mouse pointer is a double-headed arrow
    everywhere in the taskbar, which is definitely abnormal. The taskbar
    appears to be in doornail mode. So is every Explorer window, I quickly
    discover. I use alt-tab to reach the task manager where I find the CPU
    usage graph has three big gouges in it, one of which is in progress --
    one of the DC background tasks is using one core and the other core is
    IDLE. The other DC task is apparently blocking, although it's pure
    number crunching so it can't possibly be waiting for the disk, a socket,
    or just about anything else. This means the scheduler in the kernel(!)
    is malfunctioning.

    The wackiest thing of all is that the system completely recovered,
    including Explorer without requiring I restart Explorer, a short time
    later. There was no logical reason for what happened -- I'd expanded
    that same new submenu a dozen times without incident, and there's no
    reason for displaying it to require any extensive computations. If it
    does, one would hope the computations are cached. If they are, then
    expanding it for the nth time in a row should have been fast, but
    wasn't. If they aren't, then all of the expansions should have been
    equally slow. Regardless, none should have actually frozen half of the
    machine's user-interface and blocked a completely unrelated task.

    I was given to understand that the NT kernel is supposed to isolate
    application faults. So nothing Explorer did should have hung an
    unrelated distributed-computing task, especially as the latter wasn't
    even subject to I/O contention, only memory/CPU contention, and with 200
    megs of physical RAM free and one entire core idle I don't see how it
    could be memory/CPU contention either. In particular, with about 800
    megs commit charge and 1GB physical RAM the system should not have been
    thrashing, but pagefile contention is the ONLY thing I can think of that
    would hang several unrelated tasks, or any non-I/O-bound, pure-crunching
    tasks at all.

    Which means that either the system suddenly forgot it had a second
    512-stick in there and thrashed anyway (but Task Manager still reported
    1GB physical RAM during the event?) or the system needlessly thrashed
    when there was lots of free RAM, either of which indicates a serious
    kernel-level bug.

    Moreover since only CPU, memory, and disk use may have been involved
    here, I don't see video, sound, printer, or other drivers (common
    culprits) being the cause -- only the kernel itself or "core" drivers
    (HAL, keyboard, HDD, mouse, etc.) Those are all Microsoft-supplied
    drivers on the affected system and none show updates in Windows Update
    either.

    And I'd expect a bug in any of those places to cause symptoms
    frequently, and to cause blue screens. I'm not seeing anything odd,
    except the rare instance of right-click "new" hanging in Explorer, and
    this one instance of other tasks being blocked when that had happened. I
    haven't seen any blue screens whatsoever. As far as that is concerned,
    the system is rock stable.

    Aside from this disturbing incident, my primary complaints are just that
    Explorer and Firefox both leak memory, and that only Firefox can be
    restarted while preserving session state to free the leaked memory while
    avoiding a reboot; Explorer cannot be restarted with session state
    preserved except by rebooting, as far as I can determine. Worse,
    Explorer's leak seems to partially survive even a reboot. It used to
    stabilize at a size of under 50 megs after being started up with no
    preserved session state (so no already-open windows). Now if I nuke it
    from task manager (causing it to lose session state) and then reboot, it
    starts up and Explorer stabilizes at a size of over 100 megs with no
    open windows! What causes this?

    --
    There's only four things you can be certain of: taxes, change, spam, and
    death.

  2. Re: Strange hang/CPU utilization issue on dual-core XP SP2 system

    [snip explorer hang and odd scheduling behavior]

    I've had the same intermittent Explorer hand with right click "new",
    for donkey's years. I think it might even go as far back as the
    Explorer that came with the shiny new just-released Windows 95 12
    years ago. Same with Explorer incessantly leaking memory. I don't know
    of any cure, unfortunately. I do know that if a shell extension is the
    cause it's a Microsoft one.

    The scheduler acting up seems very odd. If any task was being
    completely starved while "System Idle Process" had >0% CPU then that
    task had better have been blocked waiting on I/O or sleeping or else
    the scheduler must be completely b0rked. On the other hand from your
    description it shouldn't have been blocked waiting on I/O -- you said
    it was CPU-and-RAM-bound number crunching at the time, not trying to
    report something to a server somewhere? If so it should only have
    blocked if it got swapped out, and with commit charge << physical RAM
    and the task not idle that clearly shouldn't have happened.

    I'd tend to suspect a bug in virtual memory management. Freezes and
    memory leaks can both potentially be explained by such. Memory leaks
    that survive a reboot almost cannot be explained in any other way --
    try deleting and letting Windows recreate your pagefile and see if
    Explorer has shrunk back to 50 megs after rebooting.


  3. Re: Strange hang/CPU utilization issue on dual-core XP SP2 system

    nebulous99@gmail.com wrote:
    > I've had the same intermittent Explorer hand with right click "new",
    > for donkey's years. I think it might even go as far back as the
    > Explorer that came with the shiny new just-released Windows 95 12
    > years ago. Same with Explorer incessantly leaking memory. I don't know
    > of any cure, unfortunately. I do know that if a shell extension is the
    > cause it's a Microsoft one.


    "Shell extension"?

    > I'd tend to suspect a bug in virtual memory management. Freezes and
    > memory leaks can both potentially be explained by such. Memory leaks
    > that survive a reboot almost cannot be explained in any other way --
    > try deleting and letting Windows recreate your pagefile and see if
    > Explorer has shrunk back to 50 megs after rebooting.


    Deleting the page file -- how, exactly?

    --
    There's only four things you can be certain of: taxes, change, spam, and
    death.

+ Reply to Thread