Re: Peculiar Win/11 Behavior?

I finally just hit a patch of gross discontinuities. Some inter-buffer requests are now showing as much as 3.5 s of delay, instead of the usual 4-20 ms.

It almost appears that Windows/11 is tagging some memory frames as expunge-able - write them out to paging store, and then grossly delay their read-back to continue the processing under way.

There is a steady, consistent, working set of memory pages while I’m collecting data. There is no reason for tagging them as purgeable while they are being used. But maybe Windows/11 thinks I’m being unfair to the other background processes that it wants to run?



> On Jan 17, 2025, at 03:22, David McClain <[email protected]> wrote:
> 
> Here is the historical record from last night, showing the period from around 9 pm to 11 pm when the data had higher than normal variance. Afterwards the system stabilized again, except for sporadic incidents along the way. 
> 
> The graph reads from NOW on the right, to times in the past as you move left. 
> 
> <PastedGraphic-1.png>
> 
>> On Jan 17, 2025, at 03:01, David McClain <[email protected]> wrote:
>> 
>> I am doing a long running data collection process, all written in Lisp, on a Win/11 system. 
>> 
>> I notice that there are periods in the evening when my acquisitions suffer some phase discontinuities, which cause the data to exhibit higher than normal statistical variances. And often, on the longest duration collections, the system loses phase coherence completely, requiring that I shutdown the Lisp app and restart it to clear things up. There seems to develop a permanent disjunction between the Lisp data requests and the underlying Windows I/O system. That has to be my fault, but after instrumenting the system, I have yet to find hard evidence of where that happens and when.
>> 
>> My timings between data buffer requests should all be less than 500 ms. They normally range between 4-20 ms. Last night I see in the historical record there was one instance of a delay of 486 ms - almost long enough to lose phase coherency, which seems to have happened around 11 pm local time. It is now almost 3 am local time.
>> 
>> The Lisp code is completely asynchronous, but there is nothing that should be causing isolated instances of delays 20x longer than the nominal max of 20 ms.
>> 
>> I’m running the system with Sleep deferred while data collection activities are under way. So ports going to sleep should not be a problem. The screen remains lit up the entire time and no screen saver gets instantiated.
>> 
>> So, I’m wondering if anyone else sees erratic behavior in the evenings with Windows/11? Could there be some daemon scheduled to run at night that causes I/O activity to go bonkers from a regular precise schedule? Is Windows so totally bolluxed that it is insane to be attempting scientific data collection with it?
>> 
>> (We used to have dedicated VME crates with our own processors and realtime OS to do this kind of work.)
>> 
> 


_______________________________________________
Lisp Hug - the mailing list for LispWorks users
[email protected]
http://www.lispworks.com/support/lisp-hug.html