Hi all,
I just read http://wiki.squeak.org/squeak/5682 that Tim (I think it was) mentioned, and found the part about doing alternative scheduling with a higher priority thread interesting.
The question I had was, in the example it says the high priority process can use Delay's to sleep while other processes are running, but what happens when a process does IO? Doesn't that cause the process to yield?
What I am thinking is, if I were to make a scheduler process that manages all running processes, how can he find out every situation where a process yields in some way? It would be ideal if there was a Delay millisecondsOrUntilInterupt: type message.
Thanks, Jason
_________________________________________________________________ More photos, more messages, more storageget 2GB with Windows Live Hotmail. http://imagine-windowslive.com/hotmail/?locale=en-us&ocid=TXT_TAGHM_migr...
On May 29, 2007, at 20:28 , J J wrote:
Hi all,
I just read http://wiki.squeak.org/squeak/5682 that Tim (I think it was) mentioned, and found the part about doing alternative scheduling with a higher priority thread interesting.
The question I had was, in the example it says the high priority process can use Delay's to sleep while other processes are running, but what happens when a process does IO? Doesn't that cause the process to yield?
What I am thinking is, if I were to make a scheduler process that manages all running processes, how can he find out every situation where a process yields in some way? It would be ideal if there was a Delay millisecondsOrUntilInterupt: type message.
What's the actual problem you want to solve with your alternative scheduler?
- Bert -
I just wanted to experiment with an event driven scheduler and see if it makes any kind of difference in performance, especially in the case of many concurrent processes.
From: Bert Freudenberg bert@freudenbergs.de Reply-To: The general-purpose Squeak developers listsqueak-dev@lists.squeakfoundation.org To: The general-purpose Squeak developers listsqueak-dev@lists.squeakfoundation.org Subject: Re: Alternative Thread Schedulers Date: Tue, 29 May 2007 21:10:00 +0200
On May 29, 2007, at 20:28 , J J wrote:
Hi all,
I just read http://wiki.squeak.org/squeak/5682 that Tim (I think it was) mentioned, and found the part about doing alternative scheduling with a higher priority thread interesting.
The question I had was, in the example it says the high priority process can use Delay's to sleep while other processes are running, but what happens when a process does IO? Doesn't that cause the process to yield?
What I am thinking is, if I were to make a scheduler process that manages all running processes, how can he find out every situation where a process yields in some way? It would be ideal if there was a Delay millisecondsOrUntilInterupt: type message.
What's the actual problem you want to solve with your alternative scheduler?
- Bert -
_________________________________________________________________ PC Magazines 2007 editors choice for best Web mailaward-winning Windows Live Hotmail. http://imagine-windowslive.com/hotmail/?locale=en-us&ocid=TXT_TAGHM_migr...
On 29-May-07, at 11:28 AM, J J wrote:
The question I had was, in the example it says the high priority process can use Delay's to sleep while other processes are running, but what happens when a process does IO? Doesn't that cause the process to yield?
Not as such, no. *IF* you implement the io in such a way that it can be non-blocking (see the somewhat mouldy AsynchFile class for an example) then you would indeed make the relevant thread/Process yield by way of waiting on a semaphore.
What I am thinking is, if I were to make a scheduler process that manages all running processes
I think you'll find that ProcessScheduler already does this.
, how can he find out every situation where a process yields in some way? It would be ideal if there was a Delay millisecondsOrUntilInterupt: type message.
Look in the MessageName browser for messages contaiing 'wait'. There's a gazillion.
tim -- tim Rowledge; tim@rowledge.org; http://www.rowledge.org/tim Strange OpCodes: CPM: Change Programmer's Mind
From: tim Rowledge tim@rowledge.org Reply-To: The general-purpose Squeak developers listsqueak-dev@lists.squeakfoundation.org To: The general-purpose Squeak developers listsqueak-dev@lists.squeakfoundation.org Subject: Re: Alternative Thread Schedulers Date: Tue, 29 May 2007 13:48:07 -0700
I think you'll find that ProcessScheduler already does this.
I don't understand what you mean by this. What I wanted to try was to create an alternative scheduler but without replacing the existing one (at least until I'm positive it is an improvement).
Look in the MessageName browser for messages contaiing 'wait'. There's a gazillion.
I didn't see anything that would put the process to sleep for the specified amount of time or until the first yield situation (i.e. wake me on either of these conditions).
John gave a very nice write up of how the scheduling works. What I did not see was how to hook into it from Smalltalk. For example, if you call suspend, which calls primitiveSuspend the VM calls transferTo: and wakeHighestPriority but where does this come back into Smalltalk for handling?
What I want to do is from a higher priority process to take total control of the scheduling and try out an event driven style scheduler. To be more specific, I would have, to start say, 6 different priorities: 1 real-time and 5 "normal". Real-time runs any time it wants for as long as it wants, but the rest have a specific quantum. The highest normal priority might have e.g. 20ms or maybe even just 10. The lowest might have 250ms or maybe even 400 (I believe this is what sun gave their lowest priorties). Of course tunning would be required with this.
If no real-time processes want the CPU then the highest normal process is run with something like:
nextProcess resume. "I'm assuming this doesn't start him until I, the high high prio process sleep" self waitMilliseconds: (Process quantumsFor: nextProcess priority).
Then if the process yields the CPU in some way it goes up in priority. If it uses the entire quantum then it is dropped in priority. But for this to work I would need my new scheduler class to get involved any time there was some kind of context switch. It can't happen that when the wait is over my process comes alive to find a different process is current then what he expects.
So what's the point?
Well, I see people on the list always saying to use Comanche as little as possible because it is too slow, but I wonder why this is the case. Yaws (the Erlang web server, which spawns one new process per connection) beats Apache quite handily, I wonder why Comanche couldn't do the same.
I read on the Wiki that Comanche also forks a new process for each new connection, but from what I could tell it looks like the fork is at the same priority of the server. So what this would mean is if 30 clients connect right after each other, 29 fast clients and one that requires a lot of processing then the clients will connect and be serviced until the 1 long one hits, then the rest have to wait for him to finish.
With the event driven scheduler described above each thread would be started in the highest priority with the lowest quantum (since the server spends the vast majority of his life sleeping he will quickly be promoted to top normal priority). So the first quick processes connect, make their request and yield before using their quantum. The long process would run out his entire 10 or 20ms quantum and get demoted, allowing all the rest of the process to come in and be processed before he runs again.
Now the long process would actually take longer in this case, but more different clients are serviced in the same time making the server at least appear more responsive.
_________________________________________________________________ PC Magazines 2007 editors choice for best Web mailaward-winning Windows Live Hotmail. http://imagine-windowslive.com/hotmail/?locale=en-us&ocid=TXT_TAGHM_migr...
J J wrote:
What I want to do is from a higher priority process to take total control of the scheduling and try out an event driven style scheduler.
Tweak contains an example for this. It's ScriptScheduler is a custom scheduler which guarantees that each script (process) under its control is scheduled with the semantics that I defined for it (check out the Croquet SDK to see it in action).
Then if the process yields the CPU in some way it goes up in priority. If it uses the entire quantum then it is dropped in priority. But for this to work I would need my new scheduler class to get involved any time there was some kind of context switch. It can't happen that when the wait is over my process comes alive to find a different process is current then what he expects.
You'd have to do something similar to Tweak's scheduler then. You can't beat the VM into not switching processes or putting your process in control of all the other processes. What you can do however is to instruct *your* processes such that they adhere to your rules within your context.
Well, I see people on the list always saying to use Comanche as little as possible because it is too slow, but I wonder why this is the case. Yaws (the Erlang web server, which spawns one new process per connection) beats Apache quite handily, I wonder why Comanche couldn't do the same.
I'd think because of I/O speed. It would be trivial to fire off a couple thousand processes in Squeak (I have done that). However, webservers are often not limited by how quickly they switch between processes but rather how effectively they can pump I/O (files) to the network (which, if done right, doesn't require any user-level code in the middle at all).
I read on the Wiki that Comanche also forks a new process for each new connection, but from what I could tell it looks like the fork is at the same priority of the server. So what this would mean is if 30 clients connect right after each other, 29 fast clients and one that requires a lot of processing then the clients will connect and be serviced until the 1 long one hits, then the rest have to wait for him to finish.
I haven't looked at the code but I somewhat doubt that. It'd be trivial to fix (see below).
With the event driven scheduler described above each thread would be started in the highest priority with the lowest quantum (since the server spends the vast majority of his life sleeping he will quickly be promoted to top normal priority). So the first quick processes connect, make their request and yield before using their quantum. The long process would run out his entire 10 or 20ms quantum and get demoted, allowing all the rest of the process to come in and be processed before he runs again.
Now the long process would actually take longer in this case, but more different clients are serviced in the same time making the server at least appear more responsive.
Yes, prioritizing the quick requests is certainly a good idea for something like a web-server. Though, a simpler version would be to run a high priority process that simply yields every 50msecs or so to shuffle the worker processes a little. It wouldn't be quite as effective as your event scheduler but -having done a custom scheduler- it is *infinitely* easier to implement ;-)
Cheers, - Andreas
Hi!
J J wrote:
I read on the Wiki that Comanche also forks a new process for each new connection, but from what I could tell it looks like the fork is at the same priority of the server. So what this would mean is if 30 clients connect right after each other, 29 fast clients and one that requires a lot of processing then the clients will connect and be serviced until the 1 long one hits, then the rest have to wait for him to finish.
I haven't looked at the code but I somewhat doubt that. It'd be trivial to fix (see below).
I just looked and yes, AFAICT from a visual short inspection the forked process to serve a new connection does indeed not get any specific prio - it will thus use the same as the TcpService uses, which I think is by default #userBackgroundPriority (30).
Now, I am not sure if this is a problem in practice - perhaps we already have some higher prio process that causes these processes to shuffle around as Andreas describes it.
regards, Göran
From: Andreas Raab andreas.raab@gmx.de Reply-To: The general-purpose Squeak developers listsqueak-dev@lists.squeakfoundation.org To: The general-purpose Squeak developers listsqueak-dev@lists.squeakfoundation.org Subject: Re: Alternative Thread Schedulers Date: Tue, 29 May 2007 23:41:33 -0700
Tweak contains an example for this. It's ScriptScheduler is a custom scheduler which guarantees that each script (process) under its control is scheduled with the semantics that I defined for it (check out the Croquet SDK to see it in action).
Ok, I will go check that out then.
I'd think because of I/O speed. It would be trivial to fire off a couple thousand processes in Squeak (I have done that). However, webservers are often not limited by how quickly they switch between processes but rather how effectively they can pump I/O (files) to the network (which, if done right, doesn't require any user-level code in the middle at all).
Yea, I realize the majority of the time isn't spend scheduling, but I do think such an approach could make the system feel more snappy in cases without the user having to do anything extra.
As far as the IO, maybe we could add a primitive for that system call lighttp (the fastest page serving web server on the planet afaik) uses to write files directly to the network.
Yes, prioritizing the quick requests is certainly a good idea for something like a web-server. Though, a simpler version would be to run a high priority process that simply yields every 50msecs or so to shuffle the worker processes a little. It wouldn't be quite as effective as your event scheduler but -having done a custom scheduler- it is *infinitely* easier to implement ;-)
Cheers,
- Andreas
Well I appreciate the info. If I do this it would be an experiment type thing for some ideas I have, so it's ok if it's hard to do and just gets thrown away after I do my tests. Though the harder it is to do the further down the "todo list" it goes. :)
_________________________________________________________________ More photos, more messages, more storageget 2GB with Windows Live Hotmail. http://imagine-windowslive.com/hotmail/?locale=en-us&ocid=TXT_TAGHM_migr...
On May 29, 2007, at 11:08 PM, J J wrote:
John gave a very nice write up of how the scheduling works. What I did not see was how to hook into it from Smalltalk. For example, if you call suspend, which calls primitiveSuspend the VM calls transferTo: and wakeHighestPriority but where does this come back into Smalltalk for handling?
wakeHigestPriority looks at the list of runnable processes which is a sorted list by priority. Each element in the list is actually a linked list of processes that are able to run at that priority level. There is an idle loop process at the bottom which when run means there is no work to be done by the VM. The idle task invokes a VM primitive to make the VM sleep. In the far past this was a set amount of milliseconds, years ago it was changed to sleep upto the next Delay timing point, and/or wake if the operating system received an I/ O interrupt of some form assuming the operating system and sleep call supports that.
What's important is what is the next runnable process. That can be adjusted in the ProcessScheduler Smalltalk class by suspending processes you don't want to run yet, or adjusting priorities to sort the processes into something different.
-- ======================================================================== === John M. McIntosh johnmci@smalltalkconsulting.com Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ======================================================================== ===
From: John M McIntosh johnmci@smalltalkconsulting.com Reply-To: johnmci@smalltalkconsulting.com,The general-purpose Squeak developers listsqueak-dev@lists.squeakfoundation.org To: The general-purpose Squeak developers listsqueak-dev@lists.squeakfoundation.org Subject: Re: Alternative Thread Schedulers Date: Wed, 30 May 2007 10:02:27 -0700
On May 29, 2007, at 11:08 PM, J J wrote:
What's important is what is the next runnable process. That can be adjusted in the ProcessScheduler Smalltalk class by suspending processes you don't want to run yet, or adjusting priorities to sort the processes into something different.
That is a great idea! I can just suspend every process except the one I want to be current, and the scheduler (I will probably have to have my own lists for the processes anyway). Now if my scheduler process calls #waitMilliseconds: 250 and the VM decides the current process needs to sleep, will he wake up my processor object even though the 250 milliseconds are not over, or he would go to the idle process at that point?
I guess it doesn't matter, I would just have to have a timer interupt class that did the #waitMilliseconds and the actual Scheduler class at a low priority to take control when the VM wants to switch to a new process. Everything else (except real-time processes) would be suspended until I explicitly resume them.
_________________________________________________________________ More photos, more messages, more storageget 2GB with Windows Live Hotmail. http://imagine-windowslive.com/hotmail/?locale=en-us&ocid=TXT_TAGHM_migr...
On 30-May-07, at 2:52 PM, J J wrote:
That is a great idea! I can just suspend every process except the one I want to be current, and the scheduler (
Last time I checked - about thirty seconds ago using the ProcessBrowser - there isn't a scheduler process. I suspect you need to do a fair bit more reading to find what is in the system already. It seems to me you're getting a bit confused about how the current scheduling works.
tim -- tim Rowledge; tim@rowledge.org; http://www.rowledge.org/tim Fractured Idiom:- VENI, VIDI, VICE - I came, I saw, I partied.
Right Tim, he would have to add one, need *something* to suspending things and fiddling with the list of runnable processes.
On May 30, 2007, at 3:24 PM, tim Rowledge wrote:
On 30-May-07, at 2:52 PM, J J wrote:
That is a great idea! I can just suspend every process except the one I want to be current, and the scheduler (
Last time I checked - about thirty seconds ago using the ProcessBrowser - there isn't a scheduler process. I suspect you need to do a fair bit more reading to find what is in the system already. It seems to me you're getting a bit confused about how the current scheduling works.
tim
tim Rowledge; tim@rowledge.org; http://www.rowledge.org/tim Fractured Idiom:- VENI, VIDI, VICE - I came, I saw, I partied.
-- ======================================================================== === John M. McIntosh johnmci@smalltalkconsulting.com Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ======================================================================== ===
John M McIntosh johnmci@smalltalkconsulting.com writes:
Right Tim, he would have to add one, need *something* to suspending things and fiddling with the list of runnable processes.
Yes, that should be enough. This strategy applies to all kinds of thread systems, not just Squeak's, and allows you to bootstrap a new scheduler on top of an existing one.
Some general thoughts about this scheme:
1. Make the scheduler high-priority, so that it reliably gets control whenever it wants to.
2. Possible make your own "idle" process, one notch above the current idle process, so that you can detect a lack of activity. This process can wake up the high-priority one, if you want, by signalling a semaphore that the high-priority one notices.
3. Consider limiting your schedule to managing a list of threads that are registered with it, as opposed to literally managing all threads in the system. This way it plays a little nicer with anything else that might be running.
Oh, and the obvious:
4. Save often, and make lots of backups. Screwing up thread scheduling can really ruin an image!
Lex
On Jun 1, 2007, at 7:54 AM, Lex Spoon wrote:
- Possible make your own "idle" process, one notch above the current idle process, so that you can detect a lack of activity. This process can wake up the high-priority one, if you want, by signalling a semaphore that the high-priority one notices.
Ah, this is good. Don't fiddle with the system's idle process. Many years ago I tried that, we even pushed the change out, followed a few hours later by an emergency fix. The intent was to change the idle logic from sleeping a default of 10 ms (or something) to sleeping up to the next wakeup Delay timing point.
We discovered (Scott did I think) that the code in Smalltalk which went off to find out that information *could* if the objects aligned right then wait on the Delay semeaphore. Which then result in the VM terminating because there was no runnable process since the idle process was stalled waiting on the Delay semaphore.
A few years after this we did put back in logic in the VM idle primitive to get the next wakeup time because the VM was handed that information as part of the Delay logic. Thus it can sleep upto the next wakeup tick, baring interrupts from the operating system which terminate the operating system's sleep api logic early. That improved Delay accuracy and reduced CPU needs for idle system.
-- ======================================================================== === John M. McIntosh johnmci@smalltalkconsulting.com Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ======================================================================== ===
J J writes:
What I want to do is from a higher priority process to take total control of the scheduling and try out an event driven style scheduler. To be more specific, I would have, to start say, 6 different priorities: 1 real-time and 5 "normal". Real-time runs any time it wants for as long as it wants, but the rest have a specific quantum. The highest normal priority might have e.g. 20ms or maybe even just 10. The lowest might have 250ms or maybe even 400 (I believe this is what sun gave their lowest priorties). Of course tunning would be required with this.
At some point it would be nice to have some scheduling support for Exupery. When running as a background compiler Exupery runs in a separate, Squeak, process. Ideally, Exupery should be allowed all the time when nothing else is going on but not starved completely if the process is busy so the system can still optimize itself under load.
Exupery shouldn't be able to starve other processes or get starved itself.
Bryce
From: bryce@kampjes.demon.co.uk Reply-To: The general-purpose Squeak developers listsqueak-dev@lists.squeakfoundation.org To: The general-purpose Squeak developers listsqueak-dev@lists.squeakfoundation.org Subject: Re: Alternative Thread Schedulers Date: Wed, 30 May 2007 21:20:50 +0100
At some point it would be nice to have some scheduling support for Exupery. When running as a background compiler Exupery runs in a separate, Squeak, process. Ideally, Exupery should be allowed all the time when nothing else is going on but not starved completely if the process is busy so the system can still optimize itself under load.
Exupery shouldn't be able to starve other processes or get starved itself.
Bryce
I think the scheduler I described would do this. Theoretically it would be possible for the high priority processes to keep using the CPU so that Exupery never gets to run, but in practice I think the processes will either stay off the CPU (which is why they are high priority) or if they do use the processor too much they get demoted until they are at the same priority of Exupery, at which point he can run.
_________________________________________________________________ Like the way Microsoft Office Outlook works? Youll love Windows Live Hotmail. http://imagine-windowslive.com/hotmail/?locale=en-us&ocid=TXT_TAGHM_migr...
From: tim Rowledge tim@rowledge.org
Reply-To: The general-purpose Squeak developers listsqueak-dev@lists.squeakfoundation.org To: The general-purpose Squeak developers listsqueak-dev@lists.squeakfoundation.org Subject: Re: Alternative Thread Schedulers Date: Tue, 29 May 2007 13:48:07 -0700
Not as such, no. *IF* you implement the io in such a way that it can be non-blocking (see the somewhat mouldy AsynchFile class for an example) then you would indeed make the relevant thread/Process yield by way of waiting on a semaphore.
Oh sorry, missed this part in my response. Actually, this is exactly what I *don't* want. I don't want to jump through special hoops doing IO just to play nice with scheduling, I want it to be that way by default like in the OS. Sounds like I may have more work cut out for me then I thought. :)
_________________________________________________________________ Catch suspicious messages before you open themwith Windows Live Hotmail. http://imagine-windowslive.com/hotmail/?locale=en-us&ocid=TXT_TAGHM_migr...
On May 30, 2007, at 8:21 , J J wrote:
From: tim Rowledge tim@rowledge.org
Reply-To: The general-purpose Squeak developers list<squeak- dev@lists.squeakfoundation.org> To: The general-purpose Squeak developers list<squeak- dev@lists.squeakfoundation.org> Subject: Re: Alternative Thread Schedulers Date: Tue, 29 May 2007 13:48:07 -0700
Not as such, no. *IF* you implement the io in such a way that it can be non-blocking (see the somewhat mouldy AsynchFile class for an example) then you would indeed make the relevant thread/ Process yield by way of waiting on a semaphore.
Oh sorry, missed this part in my response. Actually, this is exactly what I *don't* want. I don't want to jump through special hoops doing IO just to play nice with scheduling, I want it to be that way by default like in the OS. Sounds like I may have more work cut out for me then I thought. :)
This referred to file i/o. If I am not mistaken, socket i/o actually uses delays and semaphores a lot.
- Bert -
I'm sure there was mail earlier asking about when a process switch happens. Ok let's clarify that. There is only one place in the VM where that happens is.
Interpreter>>transferTo:
called by
primitiveSuspend -> transfer to wakeHighestPriority Called via Smalltalk Process>>suspend, Process>>terminate
primitiveWait -> transfer to wakeHighestPriority, unless excessSignals is > 0, in that case we subtract one, and don't do the transferTo: primitiveYield -> transfer to wakeHighestPriority, and stick current process on last link of queue in it's priority group. resume: -> transer to given process if it has higher priority than current process, and put current process toSleep:. Otherwise if same or less prioirty put given process toSleep:
Called on a Semaphore signal via Interpreter>>synchronousSignal: This ensures if the process is higher priority it will become un- suspend and run, or become un-suspended and be put on queue to run some day later...
I'll note the priority queue is a sorted list of linked lists.
Helper methods
toSleep: -> sticks process on last link of priority queue.
wakeHighestPriority -> fetchs the process that is runable that is the highest priority. Ending of course with the lowest priority idle loop process which is always runable. Failure to fine any runnable process terminates the VM with an error.
Semaphores
There are semaphores that are signaled via checkForInterrupts which we mentioned earlier that attempts to run every 1ms.
special semaphores are: lowSpace Signal if space is low interruptPending Signal if keyboard interrupt was done TheFinalizationSemaphore, Signal if any GC Finalization needs to run.
The other two interesting semaphores are:
ExternalSemaphores, dual queue of semaphores set usually by plugins that want to signal a semaphore. SocketPlugin is heavy user of this logic. The reason for two queues is one queue is active, the other is being processed, this avoids (one hopes) a race condition between signalling and processing because the VM has no way to lock access to the queue with a thread safe access method since that level of processor instruction support is not in the basic VM.
TheTimerSemaphore, The Delay logic tracks when the next Delay will fire, it feeds that information to the VM, checkForInterrupts by waking up every 1ms then looks at the request time, versus ms time, and decides if it needs to signal the TheTimerSemaphore, which then waits up the process waiting on the Delay.
CheckForInterrupts is run when
a) A method starts to run b) When we do a long unconditional jump bytecode (#160-167) backwards.
However that is moderated by interruptCheckCounter which tries to ensure not every call will result in a checkForInterrupts call. As earlier discussed over the years interruptCheckCounterFeedBackReset tries to moderate the setting of interruptCheckCounter to ensure for example a do:while: which would result in two (if not more) checkForInterrupt calls on each loop, the fiddling with interruptCheckCounter will try to ensure checkForInterrupt is only called every 1ms.
We use the forceInterruptCheck to force a checkForInterrupt call, that is done on finalization signaling, lowspace, become:, incremental/full GC, exernal semaphore signalling, plus a few others.
On May 29, 2007, at 11:28 AM, J J wrote:
Hi all,
I just read http://wiki.squeak.org/squeak/5682 that Tim (I think it was) mentioned, and found the part about doing alternative scheduling with a higher priority thread interesting.
The question I had was, in the example it says the high priority process can use Delay's to sleep while other processes are running, but what happens when a process does IO? Doesn't that cause the process to yield?
What I am thinking is, if I were to make a scheduler process that manages all running processes, how can he find out every situation where a process yields in some way? It would be ideal if there was a Delay millisecondsOrUntilInterupt: type message.
Thanks, Jason
More photos, more messages, more storage—get 2GB with Windows Live Hotmail. http://imagine-windowslive.com/hotmail/?locale=en- us&ocid=TXT_TAGHM_migration_HM_mini_2G_0507
-- ======================================================================== === John M. McIntosh johnmci@smalltalkconsulting.com Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ======================================================================== ===
From: John M McIntosh johnmci@smalltalkconsulting.com Reply-To: johnmci@smalltalkconsulting.com,The general-purpose Squeak developers listsqueak-dev@lists.squeakfoundation.org To: The general-purpose Squeak developers listsqueak-dev@lists.squeakfoundation.org Subject: Re: Alternative Thread Schedulers Date: Tue, 29 May 2007 14:02:16 -0700
I'm sure there was mail earlier asking about when a process switch happens.
Yea, that was me. I am still thinking about this. :)
And thanks for the write up! I will explain more my thinking in my response to Tim.
Thanks, Jason
Ok let's clarify that. There is only one place in the VM where that happens is.
Interpreter>>transferTo:
called by
primitiveSuspend -> transfer to wakeHighestPriority Called via Smalltalk Process>>suspend, Process>>terminate
primitiveWait -> transfer to wakeHighestPriority, unless excessSignals is > 0, in that case we subtract one, and don't do the transferTo: primitiveYield -> transfer to wakeHighestPriority, and stick current process on last link of queue in it's priority group. resume: -> transer to given process if it has higher priority than current process, and put current process toSleep:. Otherwise if same or less prioirty put given process toSleep:
Called on a Semaphore signal via Interpreter>>synchronousSignal: This ensures if the process is higher priority it will become un- suspend and run, or become un-suspended and be put on queue to run some day later...
I'll note the priority queue is a sorted list of linked lists.
Helper methods
toSleep: -> sticks process on last link of priority queue.
wakeHighestPriority -> fetchs the process that is runable that is the highest priority. Ending of course with the lowest priority idle loop process which is always runable. Failure to fine any runnable process terminates the VM with an error.
Semaphores
There are semaphores that are signaled via checkForInterrupts which we mentioned earlier that attempts to run every 1ms.
special semaphores are: lowSpace Signal if space is low interruptPending Signal if keyboard interrupt was done TheFinalizationSemaphore, Signal if any GC Finalization needs to run.
The other two interesting semaphores are:
ExternalSemaphores, dual queue of semaphores set usually by plugins that want to signal a semaphore. SocketPlugin is heavy user of this logic. The reason for two queues is one queue is active, the other is being processed, this avoids (one hopes) a race condition between signalling and processing because the VM has no way to lock access to the queue with a thread safe access method since that level of processor instruction support is not in the basic VM.
TheTimerSemaphore, The Delay logic tracks when the next Delay will fire, it feeds that information to the VM, checkForInterrupts by waking up every 1ms then looks at the request time, versus ms time, and decides if it needs to signal the TheTimerSemaphore, which then waits up the process waiting on the Delay.
CheckForInterrupts is run when
a) A method starts to run b) When we do a long unconditional jump bytecode (#160-167) backwards.
However that is moderated by interruptCheckCounter which tries to ensure not every call will result in a checkForInterrupts call. As earlier discussed over the years interruptCheckCounterFeedBackReset tries to moderate the setting of interruptCheckCounter to ensure for example a do:while: which would result in two (if not more) checkForInterrupt calls on each loop, the fiddling with interruptCheckCounter will try to ensure checkForInterrupt is only called every 1ms.
We use the forceInterruptCheck to force a checkForInterrupt call, that is done on finalization signaling, lowspace, become:, incremental/full GC, exernal semaphore signalling, plus a few others.
On May 29, 2007, at 11:28 AM, J J wrote:
Hi all,
I just read http://wiki.squeak.org/squeak/5682 that Tim (I think it was) mentioned, and found the part about doing alternative scheduling with a higher priority thread interesting.
The question I had was, in the example it says the high priority process can use Delay's to sleep while other processes are running, but what happens when a process does IO? Doesn't that cause the process to yield?
What I am thinking is, if I were to make a scheduler process that manages all running processes, how can he find out every situation where a process yields in some way? It would be ideal if there was a Delay millisecondsOrUntilInterupt: type message.
Thanks, Jason
More photos, more messages, more storageget 2GB with Windows Live Hotmail. http://imagine-windowslive.com/hotmail/?locale=en- us&ocid=TXT_TAGHM_migration_HM_mini_2G_0507
--
=== John M. McIntosh johnmci@smalltalkconsulting.com Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ======================================================================== ===
_________________________________________________________________ Catch suspicious messages before you open themwith Windows Live Hotmail. http://imagine-windowslive.com/hotmail/?locale=en-us&ocid=TXT_TAGHM_migr...
squeak-dev@lists.squeakfoundation.org
-
Andreas Raab -
Bert Freudenberg -
bryce＠kampjes.demon.co.uk -
Göran Krampe -
J J -
John M McIntosh -
Lex Spoon -
tim Rowledge