John M McIntosh wrote:

Well I'm not sure what exactly you are trying to fix or optimize

Sorry for not being clear. I'm trying to address several concrete problems:

1) Dragging a Squeak main window by its label currently blocks the VM. Sounds stops playing, animations stop animating etc.

2) Opening OS file dialogs or context menus block the VM. Same issues. You've had this problem in Sophie, we've added a specific solution for Teleplace and I'd like to generalize this into a generic solution.

3) Make it easy to embed a Squeak VM by making it possible to deliver an event and run the resulting code. Allowing for example to run the VM as a browser plugin by calling interpret() in response to the browser-delivered events.

Abstractly speaking, I am trying to turn around the way interpret() is being run. Instead of calling interpret() and it then calling ioProcessEvents() I want to call interpret in response to events generated elsewhere and have it return when it's done handling the events up to this point. This addresses all the issues mentioned above.

Now as for the proposed changes, we would just call the ioRelinquishProcessorForMicroseconds() logic after the interpreter() ends. On register rich machines (ancient powerpc machines this change would be technically more expensive since it loads up 20 some registers on the entry to interpret() but give it's call rate of 50 times a second no-one will notice.

For example. I am *not* proposing to implement the "trivial client loop" that I was talking about earlier. Rather, what I would do on Windows is call interpret() from the appropriate WNDPROC in response to a UI event. The main loop would be a "standard windows event loop" without any references to interpret(). I would expect that the other platforms do whatever is appropriate, the trivial client loop was intended purely as an example.

As a consequence of the changes, you would be able to have other main loops (via MFC, Cocoa, wxWidgets etc. you name it) and call interpret() in response to incoming events.

Cheers, - Andreas

John M McIntosh wrote:

Ok, well

So does that mean you are: [ ] in favor, [ ] against such a change, [ ] don't care, [ ] think it can't work at all? I'm not clear about your reply you seem to be saying all of the above at some point or other.

Cheers, - Andreas

On 2009-11-10, at 12:03 PM, Andreas Raab wrote:

...

John M McIntosh wrote:

...

Well I'm not sure what exactly you are trying to fix or optimize

Sorry for not being clear. I'm trying to address several concrete problems:

1. Dragging a Squeak main window by its label currently blocks the VM.

Sounds stops playing, animations stop animating etc.

2. Opening OS file dialogs or context menus block the VM. Same issues.

You've had this problem in Sophie, we've added a specific solution for Teleplace and I'd like to generalize this into a generic solution.

These are windows issues?

Ok for os-x these problems don't go away. Why? Well the Squeak is run as a part of the main UI thread. The choice here is:

(a) Run the squeak VM from time to time as a subtask of the UI thread, in the past this gave poor performance since the overhead is high, and the setup for interp in powerpc was high. However in this model as it services a UI task like menu interaction then squeak doesn't run.

(b) Run the main UI event processing as a subtask of the Squeak thread, the current mode. This stops the VM from running as it services UI events, again the VM is calling ioProcessEvent and that is busy servicing the menu interaction. Cost is cheaper since we just ask if there are any events pending every 16ms or so. This is why when on os-x when the morphic task is dead then the VM slowly responds to window/menu interactions since there is a 1/2 second or so ioProcessEvent callback in checkForInterrupts()

So why not run the Squeak VM as a separate thread from the UI Thread, yes we do that for the iPhone. However any UI interacts from the VM has to run on the UI thread, fine you can work with that. Now add in desire to run Objective-C methods, easy to deadlock between UI Thread callbacks and Squeak thread running message send on UI Thread. Besides all that hassle this applies to FFI calls where when building Sophie we discovered it was just illegal to call Quicktime routines from a background thread. Starts to get complicated, lets add in the fact you cann't even create an object in a background thread for use on the foreground thread because UI objects have semaphores which are thread aware and a cross thread lock isn't allowed.

...

3. Make it easy to embed a Squeak VM by making it possible to deliver

an event and run the resulting code. Allowing for example to run the VM as a browser plugin by calling interpret() in response to the browser-delivered events.

Been there done that, I think the code is still lurking, something about asking if we should terminate the interp() loop for the browser to run.

--

John M. McIntosh johnmci@smalltalkconsulting.com Twitter: squeaker68882 Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ===========================================================================

On Tue, Nov 10, 2009 at 5:56 PM, John M McIntosh < johnmci@smalltalkconsulting.com> wrote:

On 2009-11-10, at 4:35 PM, Andreas Raab wrote:

John M McIntosh wrote:

...

...

Ok, well

So does that mean you are: [ ] in favor, [ ] against such a change, [ ] don't care, [ ] think it can't work at all? I'm not clear about your reply you seem to be saying all of the above at some point or other.

Er, all of the above, from early practice on slow machines it gave sluggish response to early forms of the browser plugin.

However if you feel this will fix issues on the Windows platform then it's not a problem for the os-x and iphone version to work with it. For the iPhone perhaps we flip the model to call interpret() for some set time on each pass thru the runloop.

Current we have

#define browserPluginReturnIfNeeded() if (plugInTimeToReturn()) {ReturnFromInterpret();} #define browserPluginInitialiseIfNeeded()

plugInTimeToReturn on os-x is defined to check flags that are set if the VM is running as a sub-process of the browser and the browser has terminated, thus we want to terminate the VM.

So I'm wondering what the check will be, how many times per second will it be executed, and will interpret() run for a set number of milliseconds, a suggested amount, or a number of bytes codes? Or if I read correctly we exit from interpret() when there are no processes to run, then go back in when we think an external event happens, or a Delay terminates.

The latter. The ideal thing for the VM to do when it discovers it has no processes to run is to call select, poll et al, with a timeout determined from the pending delay. This the VM will wake up as soon as it has work to do, either because input is available or a delay has expired. This is more difficult on platforms like Windows where there is no common wait-for-input mechanism other than increasingly obsolete WaitNextEvent style interfaces (the problem being using an event-driven non-blocking socket interface, yuck).

With the threaded Squeak VM I'm working on one can go one better and have a number of image-level processes that block in the FFI and a number of worker threads in the VM that block on OS semaphores waiting for the VM to give them something to do.

Andreas, note that my threaded work already arranges to exit the interpreter/JIT combination wen a thread has nothing to do. We should coordinate changes so that we maintain compatibility.

Re callbacks, an approach like Alien's is a good one. I'm not quite at the stage with the threaded FFI work to be implementing callbacks, but it is on my list and the Alien approach is a general one.

--

=========================================================================== John M. McIntosh johnmci@smalltalkconsulting.com Twitter: squeaker68882 Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ===========================================================================

On Tue, Nov 10, 2009 at 6:45 PM, John M McIntosh < johnmci@smalltalkconsulting.com> wrote:

On 2009-11-10, at 6:17 PM, Eliot Miranda wrote:

With the threaded Squeak VM I'm working on one can go one better and have

...

a number of image-level processes that block in the FFI and a number of worker threads in the VM that block on OS semaphores waiting for the VM to give them something to do.

Obviously now you have to give a bit more details on this. Is it like the hydra VM? Or entirely different?

Orthogonal, in that it might work well with Hydra. The basic scheme is to have a natively multi-threaded VM that is not concurrent. Multiple native threads share the Vm such that there is only one thread running Vm code at any one time. This the VM can make non-blocking calls to the outside world but neither the VM nor the image need to be modified to handle true concurrency. This is the same basic architecture as in the Strongtalk and V8 VMs and notably in David Simmons' various Smalltalk VMs.

The cool thing about the system is David's design. He's been extremely generous in explaining to me his scheme, which is extremely efficient. I've merely implemented this scheme in the context of the Cog VM. The idea is to arrange that a threaded callout is so cheap that any and all callouts can be threaded. This is done by arranging that a callout does not switch to another thread, instead the thread merely "disowns" the VM. It is the job of a background heartbeat thread to detect tat a callout is long-runnijng and that the VM has effectively blocked. The heartbeat then activates a new thread to run the VM and the new thread attempts to take ownership and will run Smalltalk code if it succeeds.

On return form a callout a thread must attempt to take ownership of the VM, and if it fails, add itself to a queue of threads waiting to take back the VM and then wait on an OS semaphore until the thread owning the VM decides to give up ownership to it.

Every VM thread has a unique index. The vmOwner variable holds the index of the owning thread or 0 if the VM is unowned. To disown the VM all a thread has to do is zero vmOwner, while remembering the value of vmOwner in a temporary. To take ownership a thread must use a low-level lock to gain exclusive access to vmOwner, and if vmOwner is zero, set it back to the thread's index, and release the lock. If it finds vmOwner is non-zero it releases the lock and enters the wanting ownership queue.

In the Cog VM the heartbeat beats at 1KHz, so any call that takes less than 0.5ms is likely to complete without the heartbeat detecting that the VM is blocked. So any and all callouts can be threaded. Quite brilliant. All the work of changing the active process when switching between threads is deferred from callout time to when a different thread takes ownership of the VM, saving the VM state for the process that surrendered the VM and installing its own.

The major wrinkle in this is that in David's VM he has a pinning garbage collector which arranges that any arguments passed out through the FFI are implicitly pinned. We don't yet have a pinning garbage collector. I do plan to do one. But in the interim one quick hack, a neat idea of Andreas', is to fail calls that attempt to pass objects in new space, allowing only old objects to be passed, and to prevent the full garbage collector from running while any threaded calls are in progress.

Having cheap non-blocking calls allows e.g. - the Hydra inter-VM channels to be implemented in Smalltalk code above the threaded FFI - socket calls to be blocking calls in the image - Smalltalk code to call select/poll/WaitForMultipleEvents

There are still plenty of sticky issues to do with e.g. identifying threads that can do specific functions, such as the UI thread, and issuing OpenGL calls from the right thread, etc, etc. But these are all doable, if potentially tricky to get right. If this kind of code does migrate from the VM innards up to the image I think that's a really good thing (tm) but one will really have to know what one is doing to get it right.

HTH eliot

--

John M. McIntosh johnmci@smalltalkconsulting.com Twitter: squeaker68882 Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com ===========================================================================

On Tue, Nov 10, 2009 at 9:59 PM, Igor Stasenko siguctua@gmail.com wrote:

2009/11/11 Eliot Miranda eliot.miranda@gmail.com:

...

On Tue, Nov 10, 2009 at 6:45 PM, John M McIntosh <

johnmci@smalltalkconsulting.com> wrote:

...

...

On 2009-11-10, at 6:17 PM, Eliot Miranda wrote:

...

With the threaded Squeak VM I'm working on one can go one better and

have a number of image-level processes that block in the FFI and a number of worker threads in the VM that block on OS semaphores waiting for the VM to give them something to do.

...

...

Obviously now you have to give a bit more details on this. Is it like

the hydra VM? Or entirely different?

...

Orthogonal, in that it might work well with Hydra. The basic scheme is

to have a natively multi-threaded VM that is not concurrent. Multiple native threads share the Vm such that there is only one thread running Vm code at any one time. This the VM can make non-blocking calls to the outside world but neither the VM nor the image need to be modified to handle true concurrency. This is the same basic architecture as in the Strongtalk and V8 VMs and notably in David Simmons' various Smalltalk VMs.

...

The cool thing about the system is David's design. He's been extremely

generous in explaining to me his scheme, which is extremely efficient. I've merely implemented this scheme in the context of the Cog VM. The idea is to arrange that a threaded callout is so cheap that any and all callouts can be threaded. This is done by arranging that a callout does not switch to another thread, instead the thread merely "disowns" the VM. It is the job of a background heartbeat thread to detect tat a callout is long-runnijng and that the VM has effectively blocked. The heartbeat then activates a new thread to run the VM and the new thread attempts to take ownership and will run Smalltalk code if it succeeds.

...

On return form a callout a thread must attempt to take ownership of the

VM, and if it fails, add itself to a queue of threads waiting to take back the VM and then wait on an OS semaphore until the thread owning the VM decides to give up ownership to it.

...

Every VM thread has a unique index. The vmOwner variable holds the index

of the owning thread or 0 if the VM is unowned. To disown the VM all a thread has to do is zero vmOwner, while remembering the value of vmOwner in a temporary. To take ownership a thread must use a low-level lock to gain exclusive access to vmOwner, and if vmOwner is zero, set it back to the thread's index, and release the lock. If it finds vmOwner is non-zero it releases the lock and enters the wanting ownership queue.

...

In the Cog VM the heartbeat beats at 1KHz, so any call that takes less

than 0.5ms is likely to complete without the heartbeat detecting that the VM is blocked. So any and all callouts can be threaded. Quite brilliant. All the work of changing the active process when switching between threads is deferred from callout time to when a different thread takes ownership of the VM, saving the VM state for the process that surrendered the VM and installing its own.

...

The major wrinkle in this is that in David's VM he has a pinning garbage

collector which arranges that any arguments passed out through the FFI are implicitly pinned. We don't yet have a pinning garbage collector. I do plan to do one. But in the interim one quick hack, a neat idea of Andreas', is to fail calls that attempt to pass objects in new space, allowing only old objects to be passed, and to prevent the full garbage collector from running while any threaded calls are in progress.

...

Having cheap non-blocking calls allows e.g.

• the Hydra inter-VM channels to be implemented in Smalltalk code above

the threaded FFI

...

• socket calls to be blocking calls in the image
• Smalltalk code to call select/poll/WaitForMultipleEvents

There are still plenty of sticky issues to do with e.g. identifying

threads that can do specific functions, such as the UI thread, and issuing OpenGL calls from the right thread, etc, etc. But these are all doable, if potentially tricky to get right. If this kind of code does migrate from the VM innards up to the image I think that's a really good thing (tm) but one will really have to know what one is doing to get it right.

...

HTH eliot

I used a mutex in Hydra (each interpreter has own mutex), so any operation, which requires synchronization should be performed only after obtaining the mutex ownership. And sure, if crafted carefully, one could release the mutex before doing an external call, and "try" get it back again after call completed. If use mutexes, provided by OS, then you don't need a heartbeat process, obviously because you can simply put wait on mutex. So, i suppose you introducing the heardbeat to minimize the overhead of using synchronization primitives provided by OS, and instead using a low-level assembly code.

Just one minor thing - you mentioned the table of threads. What if some routine creating a new thread, which get unnoticed by VM, so its not registered in the VM 'threads' table, but then such thread attempts to obtain an ownership on interpreter somehow?

This can only happen on a callback or other well-defined entry-point. At these well-defined entry-points the VM checks whether there is a tag in thread-local storage (the thread's VM index). If it is not set the VM allocates the necessary per-thread storage, assigns an index and allows the thread to continue. On return from the entry-point the VM deallocates the storage, clears the thread-local storage and returns.

About inter-image communication in Hydra. The main problem that you need to pass a buffer between heads, so you need to get a lock on a recepient, while still keeping a lock on sender interpreter. But this could lead to deadlock, if recepient in own turn attempts to do the same. So, the solution, unfortunately, is to copy buffer to C heap (using malloc().. yeah :( ), and pass an event with pointer to such buffer, which then could be handled by recepient as soon as it ready to do so, in event handling routine.

But you could connect the two with a pair of pipes, right? Then al that locking and buffer allocation is in the VM. Or rather, once you have a non-blocking FFI you can just use an OS's native stream-based inter-process communications facilities.

One more thing: socket calls to be blocking calls in the image

Assuming that VM use blocking sockets, then call will block the thread & some of the image-side process. Then hearbeat thread at some point sees that VM has no owning thread and so, allows another thread, waiting in the queue to take ownership on VM. But what if there is no such thread? There is a choice: allocate new native thread and let it continue running VM, or just ignore & skip over for the next heat beat. I'd like to hear what you choose. Because depending from direction taken, on server image, which simultaneously serves, say 100 connections you may end up either with 100 + 1 native threads, or less (fixed) number of them but with risk to unable to run any VM code until some of the blocking calls completes.

There is a simple policy that is a cap on the total number of threads the VM will allocate. below this a new thread is allocated. At the limit the VM will block. But note that the pool starts at 1 and only grows as necessary up to the cap.

I'd like to note that either of above alternatives having a quite bad

scalability potential. I'd prefer to have a pool of threads, each of them serving N connections. The size of threads pool should be 2x-3x number of processor cores on host, because making more than that will not make any real difference, since single core can serve only single native thread while others will just consume the memory resources, like address space etc.

That's very similar to my numbers too. My current default is at least two threads and no more than 32, and 2 x num processors/cores in between. But these numbers should be configurable. This is just to get started.

...

...

--

===========================================================================

...

...

John M. McIntosh johnmci@smalltalkconsulting.com Twitter:

squeaker68882

...

...

Corporate Smalltalk Consulting Ltd. http://www.smalltalkconsulting.com

===========================================================================

...

...

-- Best regards, Igor Stasenko AKA sig.

On Wed, Nov 11, 2009 at 10:20 AM, Igor Stasenko siguctua@gmail.com wrote:

2009/11/11 Eliot Miranda eliot.miranda@gmail.com:

...

On Tue, Nov 10, 2009 at 9:59 PM, Igor Stasenko siguctua@gmail.com

wrote:

...

...

2009/11/11 Eliot Miranda eliot.miranda@gmail.com:

...

On Tue, Nov 10, 2009 at 6:45 PM, John M McIntosh <

johnmci@smalltalkconsulting.com> wrote:

...

...

...

...

On 2009-11-10, at 6:17 PM, Eliot Miranda wrote:

...

With the threaded Squeak VM I'm working on one can go one better and

have a number of image-level processes that block in the FFI and a number of worker threads in the VM that block on OS semaphores waiting for the VM to give them something to do.

...

...

...

...

Obviously now you have to give a bit more details on this. Is it like

the hydra VM? Or entirely different?

...

...

...

Orthogonal, in that it might work well with Hydra. The basic scheme

is to have a natively multi-threaded VM that is not concurrent. Multiple native threads share the Vm such that there is only one thread running Vm code at any one time. This the VM can make non-blocking calls to the outside world but neither the VM nor the image need to be modified to handle true concurrency. This is the same basic architecture as in the Strongtalk and V8 VMs and notably in David Simmons' various Smalltalk VMs.

...

...

...

The cool thing about the system is David's design. He's been

extremely generous in explaining to me his scheme, which is extremely efficient. I've merely implemented this scheme in the context of the Cog VM. The idea is to arrange that a threaded callout is so cheap that any and all callouts can be threaded. This is done by arranging that a callout does not switch to another thread, instead the thread merely "disowns" the VM. It is the job of a background heartbeat thread to detect tat a callout is long-runnijng and that the VM has effectively blocked. The heartbeat then activates a new thread to run the VM and the new thread attempts to take ownership and will run Smalltalk code if it succeeds.

...

...

...

On return form a callout a thread must attempt to take ownership of

the VM, and if it fails, add itself to a queue of threads waiting to take back the VM and then wait on an OS semaphore until the thread owning the VM decides to give up ownership to it.

...

...

...

Every VM thread has a unique index. The vmOwner variable holds the

index of the owning thread or 0 if the VM is unowned. To disown the VM all a thread has to do is zero vmOwner, while remembering the value of vmOwner in a temporary. To take ownership a thread must use a low-level lock to gain exclusive access to vmOwner, and if vmOwner is zero, set it back to the thread's index, and release the lock. If it finds vmOwner is non-zero it releases the lock and enters the wanting ownership queue.

...

...

...

In the Cog VM the heartbeat beats at 1KHz, so any call that takes less

than 0.5ms is likely to complete without the heartbeat detecting that the VM is blocked. So any and all callouts can be threaded. Quite brilliant. All the work of changing the active process when switching between threads is deferred from callout time to when a different thread takes ownership of the VM, saving the VM state for the process that surrendered the VM and installing its own.

...

...

...

The major wrinkle in this is that in David's VM he has a pinning

garbage collector which arranges that any arguments passed out through the FFI are implicitly pinned. We don't yet have a pinning garbage collector. I do plan to do one. But in the interim one quick hack, a neat idea of Andreas', is to fail calls that attempt to pass objects in new space, allowing only old objects to be passed, and to prevent the full garbage collector from running while any threaded calls are in progress.

...

...

...

Having cheap non-blocking calls allows e.g.

• the Hydra inter-VM channels to be implemented in Smalltalk code

above the threaded FFI

...

...

...

• socket calls to be blocking calls in the image
• Smalltalk code to call select/poll/WaitForMultipleEvents

There are still plenty of sticky issues to do with e.g. identifying

threads that can do specific functions, such as the UI thread, and issuing OpenGL calls from the right thread, etc, etc. But these are all doable, if potentially tricky to get right. If this kind of code does migrate from the VM innards up to the image I think that's a really good thing (tm) but one will really have to know what one is doing to get it right.

...

...

...

HTH eliot

I used a mutex in Hydra (each interpreter has own mutex), so any operation, which requires synchronization should be performed only after obtaining the mutex ownership. And sure, if crafted carefully, one could release the mutex before doing an external call, and "try" get it back again after call completed. If use mutexes, provided by OS, then you don't need a heartbeat process, obviously because you can simply put wait on mutex. So, i suppose you introducing the heardbeat to minimize the overhead of using synchronization primitives provided by OS, and instead using a low-level assembly code.

Just one minor thing - you mentioned the table of threads. What if some routine creating a new thread, which get unnoticed by VM, so its not registered in the VM 'threads' table, but then such thread attempts to obtain an ownership on interpreter somehow?

This can only happen on a callback or other well-defined entry-point. At

these well-defined entry-points the VM checks whether there is a tag in thread-local storage (the thread's VM index). If it is not set the VM allocates the necessary per-thread storage, assigns an index and allows the thread to continue. On return from the entry-point the VM deallocates the storage, clears the thread-local storage and returns.

...

Yes. Just to make sure everything is ok with that :)

...

...

About inter-image communication in Hydra. The main problem that you need to pass a buffer between heads, so you need to get a lock on a recepient, while still keeping a lock on sender interpreter. But this could lead to deadlock, if recepient in own turn attempts to do the same. So, the solution, unfortunately, is to copy buffer to C heap (using malloc().. yeah :( ), and pass an event with pointer to such buffer, which then could be handled by recepient as soon as it ready to do so, in event handling routine.

But you could connect the two with a pair of pipes, right? Then al that

locking and buffer allocation is in the VM. Or rather, once you have a non-blocking FFI you can just use an OS's native stream-based inter-process communications facilities.

...

of course i could. but the task is to minimize the overhead, possibly even without buffer copy overhead (that where pinning GC would be really helpfull). i don't think that OS facilities not copying data buffer to secure location before passing it between the sockets. Because once it releases the sender, while still waiting receiver to be ready to retrieve the data, it can't guarantee that given buffer will not be used for something else, hence it inevitable should either copy buffer contents to secure location or block the sender.

OK, instead one can create a buffer from within Smalltalk (e.g. via Alien) and then create OS semaphores and use blocking calls to wait on those semaphores. All I'm really trying to say is that once you have a threaded FFI you can move lots of stuff up out of the VM. The disadvantage is that one loses platform independence, but I've long thought that Smalltalk class hierarchies are a much nicer way of creating cross-platform abstractions than great gobs iof platform-specific C code.

Andreas counters that implementing the abstractions in the VM keeps them well-defined and free from meddling. But that runs counter to the philosophy of an open system and preventing inadvertent meddling is something Smalltalk has to do anyway (e.g. "Process should not be redefined, proceed to store over it"). The nice things about shooting oneself in the foot by meddling with a Smalltalk system are that a) it doesn't really do any harm and b) the debugging of it can be a great learning experience.

...

...

One more thing: socket calls to be blocking calls in the image

Assuming that VM use blocking sockets, then call will block the thread & some of the image-side process. Then hearbeat thread at some point sees that VM has no owning thread and so, allows another thread, waiting in the queue to take ownership on VM. But what if there is no such thread? There is a choice: allocate new native thread and let it continue running VM, or just ignore & skip over for the next heat beat. I'd like to hear what you choose. Because depending from direction taken, on server image, which simultaneously serves, say 100 connections you may end up either with 100 + 1 native threads, or less (fixed) number of them but with risk to unable to run any VM code until some of the blocking calls completes.

There is a simple policy that is a cap on the total number of threads

the VM will allocate. below this a new thread is allocated. At the limit the VM will block. But note that the pool starts at 1 and only grows as necessary up to the cap.

...

...

I'd like to note that either of above alternatives having a quite bad scalability potential. I'd prefer to have a pool of threads, each of them serving N connections. The size of threads pool should be 2x-3x number of processor cores on host, because making more than that will not make any real difference, since single core can serve only single native thread while others will just consume the memory resources, like address space etc.

That's very similar to my numbers too. My current default is at least

two threads and no more than 32, and 2 x num processors/cores in between. But these numbers should be configurable. This is just to get started.

Yes, but blocking sockets won't allow you to distribute load evenly when number of threads less than number of active sockets. All active connections should be distributed evenly among worker threads, that will guarantee that you consuming computing resources optimally.

So in that case one needs more threads, and to support them one needs more memory. But it shouldn't be a surprise that one needs more resources to support a higher workload. Yer takes yer choice and yer pays yer price. And of course one can providing settings to control the per-thread stack space etc.

And what about scheduling? Have you considered my idea to move

scheduling to language side, while on VM side, leave very small portion (in amount of code & logic) for switching the active processes? I think that with introduction of JIT the overhead of language-side sheduling will be quite small and quite acceptable given that it allows us to change things whenever we want, without touching VM.

No I haven't considered this because at least in my implementation, the scheduler and the thread manager are intimately connected. Once a process is in a callout on a particular thread it is bound to that thread and if it calls back it'll run on that thread and only on that thread until the callout unwinds. Further, one will be able to bind processes to specific threads to ensure that certain activities happen from a given native thread (e.g. OpenGL and the Windows debugging API both require that calls are issued from a single thread). So the thread manager and scheduler cooperate to bind processes to threads an to arrange that the right thread switches occur when process switches require them. Figuring out how to do that with a Smalltalk-level scheduler is more than I can manage right now :)

...

...

...

...

--

===========================================================================

...

...

...

...

John M. McIntosh johnmci@smalltalkconsulting.com Twitter:

squeaker68882

...

...

...

...

Corporate Smalltalk Consulting Ltd.

http://www.smalltalkconsulting.com

...

...

...

...

===========================================================================

...

...

...

...

-- Best regards, Igor Stasenko AKA sig.

-- Best regards, Igor Stasenko AKA sig.

On Wed, Nov 11, 2009 at 10:58 AM, Andreas Raab andreas.raab@gmx.de wrote:

Eliot Miranda wrote:

...

Andreas counters that implementing the abstractions in the VM keeps them well-defined and free from meddling. But that runs counter to the philosophy of an open system and preventing inadvertent meddling is something Smalltalk has to do anyway (e.g. "Process should not be redefined, proceed to store over it"). The nice things about shooting oneself in the foot by meddling with a Smalltalk system are that a) it doesn't really do any harm and b) the debugging of it can be a great learning experience.

I disagree with both of these statements 100%. First, once you have passed the first invalid pointer to some C function you will find out the hard way that "not really doing any harm" means, oh well it'll crash your image instead of raising a debugger. The consequence of this is that for any serious development you will no longer want to do in-image development; consider writing C code inside the running C app and then have it core-dump every time you misspell something. Not exactly my definition of "not really doing any harm". Secondly, wading through gobs of platform specific code is only a great learning experience if you are trying to learn platform specific stuff. Otherwise it's a useless distraction that only gets into your way of seeing (and using) the abstractions.

I'm saying the opposite. I'm saying one should implement these cross-platform abstractions up in the image. I'm saying that shooting one's self in the foot in Smalltalk is doing something like Smalltalk := nil and learning from the result.

The crash on misspellings is a straw man. We've both been in the situation where we've had to debug VM crashes, for a number of reasons, both VM bugs and invalid FFI calls. None of that has made us any the less keen on in-image development.

Attempting to implement cross-platform abstractions is a good learning experience if that's what you're into. Implementing them in a high-level language with good facilities is IMO better than implementing them in C in a context where its very hard to improve them because one can't easily experiment.

If one is in the situation of causing crashes in FFI calls one can always run the VM under a debugger, in which case one is in much the same situation as running in C. It's not much worse. And an FFI can use e.g. exception handling around a call to catch crashes and report them back to the image without exiting the image. This isn't particularly helpful because there's no stack backtrace, simply an exception code, but it at least allows the system to report the bug rather than just crash. To debug it one either has to scratch one's head or fire up that low-level debugger. Giving up on image-based development is a rather extreme reaction.

best Eliot

Cheers,

• Andreas

2009/11/11 Eliot Miranda eliot.miranda@gmail.com:

On Wed, Nov 11, 2009 at 10:20 AM, Igor Stasenko siguctua@gmail.com wrote:

...

2009/11/11 Eliot Miranda eliot.miranda@gmail.com:

...

On Tue, Nov 10, 2009 at 9:59 PM, Igor Stasenko siguctua@gmail.com wrote:

...

2009/11/11 Eliot Miranda eliot.miranda@gmail.com:

...

On Tue, Nov 10, 2009 at 6:45 PM, John M McIntosh johnmci@smalltalkconsulting.com wrote:

...

On 2009-11-10, at 6:17 PM, Eliot Miranda wrote:

> With the threaded Squeak VM I'm working on one can go one better and have a number of image-level processes that block in the FFI and a number of worker threads in the VM that block on OS semaphores waiting for the VM to give them something to do.

Obviously now you have to give a bit more details on this. Is it like the hydra VM? Or entirely different?

Orthogonal, in that it might work well with Hydra.  The basic scheme is to have a natively multi-threaded VM that is not concurrent.  Multiple native threads share the Vm such that there is only one thread running Vm code at any one time.  This the VM can make non-blocking calls to the outside world but neither the VM nor the image need to be modified to handle true concurrency.  This is the same basic architecture as in the Strongtalk and V8 VMs and notably in David Simmons' various Smalltalk VMs. The cool thing about the system is David's design.  He's been extremely generous in explaining to me his scheme, which is extremely efficient.  I've merely implemented this scheme in the context of the Cog VM.  The idea is to arrange that a threaded callout is so cheap that any and all callouts can be threaded.  This is done by arranging that a callout does not switch to another thread, instead the thread merely "disowns" the VM.  It is the job of a background heartbeat thread to detect tat a callout is long-runnijng and that the VM has effectively blocked.  The heartbeat then activates a new thread to run the VM and the new thread attempts to take ownership and will run Smalltalk code if it succeeds. On return form a callout a thread must attempt to take ownership of the VM, and if it fails, add itself to a queue of threads waiting to take back the VM and then wait on an OS semaphore until the thread owning the VM decides to give up ownership to it. Every VM thread has a unique index.  The vmOwner variable holds the index of the owning thread or 0 if the VM is unowned.  To disown the VM all a thread has to do is zero vmOwner, while remembering the value of vmOwner in a temporary.  To take ownership a thread must use a low-level lock to gain exclusive access to vmOwner, and if vmOwner is zero, set it back to the thread's index, and release the lock.  If it finds vmOwner is non-zero it releases the lock and enters the wanting ownership queue. In the Cog VM the heartbeat beats at 1KHz, so any call that takes less than 0.5ms is likely to complete without the heartbeat detecting that the VM is blocked.  So any and all callouts can be threaded.  Quite brilliant.  All the work of changing the active process when switching between threads is deferred from callout time to when a different thread takes ownership of the VM, saving the VM state for the process that surrendered the VM and installing its own. The major wrinkle in this is that in David's VM he has a pinning garbage collector which arranges that any arguments passed out through the FFI are implicitly pinned.  We don't yet have a pinning garbage collector.  I do plan to do one.  But in the interim one quick hack, a neat idea of Andreas', is to fail calls that attempt to pass objects in new space, allowing only old objects to be passed, and to prevent the full garbage collector from running while any threaded calls are in progress. Having cheap non-blocking calls allows e.g.

• the Hydra inter-VM channels to be implemented in Smalltalk code above the threaded FFI
• socket calls to be blocking calls in the image
• Smalltalk code to call select/poll/WaitForMultipleEvents

There are still plenty of sticky issues to do with e.g. identifying threads that can do specific functions, such as the UI thread, and issuing OpenGL calls from the right thread, etc, etc.  But these are all doable, if potentially tricky to get right.  If this kind of code does migrate from the VM innards up to the image I think that's a really good thing (tm) but one will really have to know what one is doing to get it right. HTH eliot

I used a mutex in Hydra (each interpreter has own mutex), so any operation, which requires synchronization should be performed only after obtaining the mutex ownership. And sure, if crafted carefully, one could release the mutex before doing an external call, and "try" get it back again after call completed. If use mutexes, provided by OS, then you don't need a heartbeat process, obviously because you can simply put wait on mutex. So, i suppose you introducing the heardbeat to minimize the overhead of using synchronization primitives provided by OS, and instead using a low-level assembly code.

Just one minor thing - you mentioned the table of threads. What if some routine creating a new thread, which get unnoticed by VM, so its not registered in the VM 'threads' table,  but then such thread attempts to obtain an ownership on interpreter somehow?

This can only happen on a callback or other well-defined entry-point.  At these well-defined entry-points the VM checks whether there is a tag in thread-local storage (the thread's VM index).  If it is not set the VM allocates the necessary per-thread storage, assigns an index and allows the thread to continue.  On return from the entry-point the VM deallocates the storage, clears the thread-local storage and returns.

Yes. Just to make sure everything is ok with that :)

...

...

About inter-image communication in Hydra. The main problem that you need to pass a buffer between heads, so you need to get a lock on a recepient, while still keeping a lock on sender interpreter. But this could lead to deadlock, if recepient in own turn attempts to do the same. So, the solution, unfortunately, is to copy buffer to C heap (using malloc().. yeah :( ), and pass an event with pointer to such buffer, which then could be handled by recepient as soon as it ready to do so, in event handling routine.

But you could connect the two with a pair of pipes, right?  Then al that locking and buffer allocation is in the VM.  Or rather, once you have a non-blocking FFI you can just use an OS's native stream-based inter-process communications facilities.

of course i could. but the task is to minimize the overhead, possibly even without buffer copy overhead (that where pinning GC would be really helpfull). i don't think that OS facilities not copying data buffer to secure location before passing it between the sockets. Because once it releases the sender, while still waiting receiver to be ready to retrieve the data, it can't guarantee that given buffer will not be used for something else, hence it inevitable should either copy buffer contents to secure location or block the sender.

OK, instead one can create a buffer from within Smalltalk (e.g. via Alien) and then create OS semaphores and use blocking calls to wait on those semaphores.  All I'm really trying to say is that once you have a threaded FFI you can move lots of stuff up out of the VM.  The disadvantage is that one loses platform independence, but I've long thought that Smalltalk class hierarchies are a much nicer way of creating cross-platform abstractions than great gobs iof platform-specific C code. Andreas counters that implementing the abstractions in the VM keeps them well-defined and free from meddling.  But that runs counter to the philosophy of an open system and preventing inadvertent meddling is something Smalltalk has to do anyway  (e.g. "Process should not be redefined, proceed to store over it").  The nice things about shooting oneself in the foot by meddling with a Smalltalk system are that a) it doesn't really do any harm and b) the debugging of it can be a great learning experience.

...

...

...

One more thing:  socket calls to be blocking calls in the image

Assuming that VM use blocking sockets, then call will block the thread & some of the image-side process. Then hearbeat thread at some point sees that VM has no owning thread and so, allows another thread, waiting in the queue to take ownership on VM. But what if there is no such thread? There is a choice: allocate new native thread and let it continue running VM, or just ignore &  skip over for the next heat beat. I'd like to hear what you choose. Because depending from direction taken, on server image, which simultaneously serves, say 100 connections you may end up either with 100 + 1 native threads, or less (fixed) number of them but with risk to unable to run any VM code until some of the blocking calls completes.

There is a simple policy that is a cap on the total number of threads the VM will allocate.  below this a new thread is allocated.  At the limit the VM will block.  But note that the pool starts at 1 and only grows as necessary up to the cap.

...

I'd like to note that either of above alternatives having a quite bad scalability potential. I'd prefer to have a pool of threads, each of them serving N connections. The size of threads pool should be 2x-3x number of processor cores on host, because making more than that will not make any real difference, since single core can serve only single native thread while others will just consume the memory resources, like address space etc.

That's very similar to my numbers too.  My current default is at least two threads and no more than 32, and 2 x num processors/cores in between.  But these numbers should be configurable.  This is just to get started.

Yes, but blocking sockets won't allow you to distribute load evenly when number of threads less than number of active sockets. All active connections should be distributed evenly among worker threads, that will guarantee that you consuming computing resources optimally.

So in that case one needs more threads, and to support them one needs more memory.  But it shouldn't be a surprise that one needs more resources to support a higher workload. Yer takes yer choice and yer pays yer price.  And of course one can providing settings to control the per-thread stack space etc.

...

And what about scheduling? Have you considered my idea to move scheduling to language side, while on VM side, leave very small portion (in amount of code & logic) for switching the active processes? I think that with introduction of JIT the overhead of language-side sheduling will be quite small and quite acceptable given that it allows us to change things whenever we want, without touching VM.

No I haven't considered this because at least in my implementation, the scheduler and the thread manager are intimately connected.  Once a process is in a callout on a particular thread it is bound to that thread and if it calls back it'll run on that thread and only on that thread until the callout unwinds.  Further, one will be able to bind processes to specific threads to ensure that certain activities happen from a given native thread (e.g. OpenGL and the Windows debugging API both require that calls are issued from a single thread).  So the thread manager and scheduler cooperate to bind processes to threads an to arrange that the right thread switches occur when process switches require them.  Figuring out how to do that with a Smalltalk-level scheduler is more than I can manage right now :)

Hmm.. i'm not sure that binding a process to thread will guarantee that all callouts will be made from same thread. Consider the code (presumably using some external API, which should be used only from main thread), which runs within the callback:

[ gl doSomething ] fork

suppose that active process is bound to particular native thread, but the process which is forked - not. But the problem that you must also bind a forked process to be run using same thread as a process which created it otherwise you having virtually no guarantees that some call will be made using wrong thread. But you can't predict what forked process does, it may do wrong calls or may not , so binding the forked process to same thread is very pessimistic choice.

Instead, why not expose the new VM abilities to language, so one could tell that specified callout should use a specified thread. Something like:

threadHandle := Smalltalk currentThreadId.

externalFunction call: { arguments } inThread: threadHandle

while by default

externalFunction call: { arguments }

will be free to run in any thread which is controlled by VM.

Then you are freed from implementing a complex and quite fragile (as to me) logic on VM side which magically attempts to keep all horses full-fed :)

...

2009/11/11 Eliot Miranda eliot.miranda@gmail.com:

On Wed, Nov 11, 2009 at 11:16 AM, Igor Stasenko siguctua@gmail.com wrote:

...

2009/11/11 Eliot Miranda eliot.miranda@gmail.com:

...

On Wed, Nov 11, 2009 at 10:20 AM, Igor Stasenko siguctua@gmail.com wrote:

...

2009/11/11 Eliot Miranda eliot.miranda@gmail.com:

...

On Tue, Nov 10, 2009 at 9:59 PM, Igor Stasenko siguctua@gmail.com wrote:

...

2009/11/11 Eliot Miranda eliot.miranda@gmail.com: > > > > On Tue, Nov 10, 2009 at 6:45 PM, John M McIntosh johnmci@smalltalkconsulting.com wrote: >> >> On 2009-11-10, at 6:17 PM, Eliot Miranda wrote: >> >>> With the threaded Squeak VM I'm working on one can go one better and have a number of image-level processes that block in the FFI and a number of worker threads in the VM that block on OS semaphores waiting for the VM to give them something to do. >> >> Obviously now you have to give a bit more details on this. Is it like the hydra VM? Or entirely different? > > Orthogonal, in that it might work well with Hydra.  The basic scheme is to have a natively multi-threaded VM that is not concurrent.  Multiple native threads share the Vm such that there is only one thread running Vm code at any one time.  This the VM can make non-blocking calls to the outside world but neither the VM nor the image need to be modified to handle true concurrency.  This is the same basic architecture as in the Strongtalk and V8 VMs and notably in David Simmons' various Smalltalk VMs. > The cool thing about the system is David's design.  He's been extremely generous in explaining to me his scheme, which is extremely efficient.  I've merely implemented this scheme in the context of the Cog VM.  The idea is to arrange that a threaded callout is so cheap that any and all callouts can be threaded.  This is done by arranging that a callout does not switch to another thread, instead the thread merely "disowns" the VM.  It is the job of a background heartbeat thread to detect tat a callout is long-runnijng and that the VM has effectively blocked.  The heartbeat then activates a new thread to run the VM and the new thread attempts to take ownership and will run Smalltalk code if it succeeds. > On return form a callout a thread must attempt to take ownership of the VM, and if it fails, add itself to a queue of threads waiting to take back the VM and then wait on an OS semaphore until the thread owning the VM decides to give up ownership to it. > Every VM thread has a unique index.  The vmOwner variable holds the index of the owning thread or 0 if the VM is unowned.  To disown the VM all a thread has to do is zero vmOwner, while remembering the value of vmOwner in a temporary.  To take ownership a thread must use a low-level lock to gain exclusive access to vmOwner, and if vmOwner is zero, set it back to the thread's index, and release the lock.  If it finds vmOwner is non-zero it releases the lock and enters the wanting ownership queue. > In the Cog VM the heartbeat beats at 1KHz, so any call that takes less than 0.5ms is likely to complete without the heartbeat detecting that the VM is blocked.  So any and all callouts can be threaded.  Quite brilliant.  All the work of changing the active process when switching between threads is deferred from callout time to when a different thread takes ownership of the VM, saving the VM state for the process that surrendered the VM and installing its own. > The major wrinkle in this is that in David's VM he has a pinning garbage collector which arranges that any arguments passed out through the FFI are implicitly pinned.  We don't yet have a pinning garbage collector.  I do plan to do one.  But in the interim one quick hack, a neat idea of Andreas', is to fail calls that attempt to pass objects in new space, allowing only old objects to be passed, and to prevent the full garbage collector from running while any threaded calls are in progress. > Having cheap non-blocking calls allows e.g. > - the Hydra inter-VM channels to be implemented in Smalltalk code above the threaded FFI > - socket calls to be blocking calls in the image > - Smalltalk code to call select/poll/WaitForMultipleEvents > There are still plenty of sticky issues to do with e.g. identifying threads that can do specific functions, such as the UI thread, and issuing OpenGL calls from the right thread, etc, etc.  But these are all doable, if potentially tricky to get right.  If this kind of code does migrate from the VM innards up to the image I think that's a really good thing (tm) but one will really have to know what one is doing to get it right. > HTH > eliot

I used a mutex in Hydra (each interpreter has own mutex), so any operation, which requires synchronization should be performed only after obtaining the mutex ownership. And sure, if crafted carefully, one could release the mutex before doing an external call, and "try" get it back again after call completed. If use mutexes, provided by OS, then you don't need a heartbeat process, obviously because you can simply put wait on mutex. So, i suppose you introducing the heardbeat to minimize the overhead of using synchronization primitives provided by OS, and instead using a low-level assembly code.

Just one minor thing - you mentioned the table of threads. What if some routine creating a new thread, which get unnoticed by VM, so its not registered in the VM 'threads' table,  but then such thread attempts to obtain an ownership on interpreter somehow?

This can only happen on a callback or other well-defined entry-point.  At these well-defined entry-points the VM checks whether there is a tag in thread-local storage (the thread's VM index).  If it is not set the VM allocates the necessary per-thread storage, assigns an index and allows the thread to continue.  On return from the entry-point the VM deallocates the storage, clears the thread-local storage and returns.

Yes. Just to make sure everything is ok with that :)

...

...

About inter-image communication in Hydra. The main problem that you need to pass a buffer between heads, so you need to get a lock on a recepient, while still keeping a lock on sender interpreter. But this could lead to deadlock, if recepient in own turn attempts to do the same. So, the solution, unfortunately, is to copy buffer to C heap (using malloc().. yeah :( ), and pass an event with pointer to such buffer, which then could be handled by recepient as soon as it ready to do so, in event handling routine.

But you could connect the two with a pair of pipes, right?  Then al that locking and buffer allocation is in the VM.  Or rather, once you have a non-blocking FFI you can just use an OS's native stream-based inter-process communications facilities.

of course i could. but the task is to minimize the overhead, possibly even without buffer copy overhead (that where pinning GC would be really helpfull). i don't think that OS facilities not copying data buffer to secure location before passing it between the sockets. Because once it releases the sender, while still waiting receiver to be ready to retrieve the data, it can't guarantee that given buffer will not be used for something else, hence it inevitable should either copy buffer contents to secure location or block the sender.

OK, instead one can create a buffer from within Smalltalk (e.g. via Alien) and then create OS semaphores and use blocking calls to wait on those semaphores.  All I'm really trying to say is that once you have a threaded FFI you can move lots of stuff up out of the VM.  The disadvantage is that one loses platform independence, but I've long thought that Smalltalk class hierarchies are a much nicer way of creating cross-platform abstractions than great gobs iof platform-specific C code. Andreas counters that implementing the abstractions in the VM keeps them well-defined and free from meddling.  But that runs counter to the philosophy of an open system and preventing inadvertent meddling is something Smalltalk has to do anyway  (e.g. "Process should not be redefined, proceed to store over it").  The nice things about shooting oneself in the foot by meddling with a Smalltalk system are that a) it doesn't really do any harm and b) the debugging of it can be a great learning experience.

...

...

...

One more thing:  socket calls to be blocking calls in the image

Assuming that VM use blocking sockets, then call will block the thread & some of the image-side process. Then hearbeat thread at some point sees that VM has no owning thread and so, allows another thread, waiting in the queue to take ownership on VM. But what if there is no such thread? There is a choice: allocate new native thread and let it continue running VM, or just ignore &  skip over for the next heat beat. I'd like to hear what you choose. Because depending from direction taken, on server image, which simultaneously serves, say 100 connections you may end up either with 100 + 1 native threads, or less (fixed) number of them but with risk to unable to run any VM code until some of the blocking calls completes.

There is a simple policy that is a cap on the total number of threads the VM will allocate.  below this a new thread is allocated.  At the limit the VM will block.  But note that the pool starts at 1 and only grows as necessary up to the cap.

...

I'd like to note that either of above alternatives having a quite bad scalability potential. I'd prefer to have a pool of threads, each of them serving N connections. The size of threads pool should be 2x-3x number of processor cores on host, because making more than that will not make any real difference, since single core can serve only single native thread while others will just consume the memory resources, like address space etc.

That's very similar to my numbers too.  My current default is at least two threads and no more than 32, and 2 x num processors/cores in between.  But these numbers should be configurable.  This is just to get started.

Yes, but blocking sockets won't allow you to distribute load evenly when number of threads less than number of active sockets. All active connections should be distributed evenly among worker threads, that will guarantee that you consuming computing resources optimally.

So in that case one needs more threads, and to support them one needs more memory.  But it shouldn't be a surprise that one needs more resources to support a higher workload. Yer takes yer choice and yer pays yer price.  And of course one can providing settings to control the per-thread stack space etc.

...

And what about scheduling? Have you considered my idea to move scheduling to language side, while on VM side, leave very small portion (in amount of code & logic) for switching the active processes? I think that with introduction of JIT the overhead of language-side sheduling will be quite small and quite acceptable given that it allows us to change things whenever we want, without touching VM.

No I haven't considered this because at least in my implementation, the scheduler and the thread manager are intimately connected.  Once a process is in a callout on a particular thread it is bound to that thread and if it calls back it'll run on that thread and only on that thread until the callout unwinds.  Further, one will be able to bind processes to specific threads to ensure that certain activities happen from a given native thread (e.g. OpenGL and the Windows debugging API both require that calls are issued from a single thread).  So the thread manager and scheduler cooperate to bind processes to threads an to arrange that the right thread switches occur when process switches require them.  Figuring out how to do that with a Smalltalk-level scheduler is more than I can manage right now :)

Hmm.. i'm not sure that binding a process to thread will guarantee that all callouts will be made from same thread. Consider the code (presumably using some external API, which should be used only from main thread), which runs within the callback:

[ gl doSomething ] fork

suppose that active process is bound to particular native thread, but the process which is forked - not. But the problem that you must also bind a forked process to be run using same thread as a process which created it otherwise you having virtually no guarantees that some call will be made using wrong thread. But you can't predict what forked process does, it may do wrong calls or may not , so binding the forked process to same thread is very pessimistic choice.

Instead, why not expose the new VM abilities to language, so one could tell that specified callout should use a specified thread. Something like:

threadHandle := Smalltalk currentThreadId.

externalFunction call: { arguments } inThread: threadHandle

while by default

externalFunction call: { arguments }

will be free to run in any thread which is controlled by VM.

Because that'll deadlock if the thread is doing something else.  Dedicating a thread to a process can avoid that.  e.g. spawn a process and bind it to a thread.  wrap the process in an API object.  The external API methods pass in requests via blocks to the hidden server process.  AFAICT Croquet uses this kind of approach to send messages between processes.

well i think that its easy to think out the solution how to tell a callout to be performed in specific thread. And i don't think that if you having such control at language level it will make any difference, because you still have to deal with it anyways, even with model which you currently employing.

...

Then you are freed from implementing a complex and quite fragile (as to me) logic on VM side which magically attempts to keep all horses full-fed :)

The mechanisms I've implemented don't look that fragile to me.  But its too early to report any real experience.  You could be right, but what I've done so far makes sense to me :)

I am right ;) , because from interpreter's point of view, switching to different Process means just switching an active context, which is a quite regular procedure for VM - interpreter switching an active context all the times at each message send and return! So, ask yourself, why VM would want to know something extra except the context it should continue interpreting from?

Potentially a cost of switching an active process == cost of message send. But of course, all logic which needs to pick what context the interpreter should switch to may introduce some overhead. But i prefer to have such logic implemented at language side, instead to be frozen inside VM.

...

...

...

-- Best regards, Igor Stasenko AKA sig.

2009/11/12 David T. Lewis lewis@mail.msen.com:

On Wed, Nov 11, 2009 at 09:16:44PM +0200, Igor Stasenko wrote:

...

Instead, why not expose the new VM abilities to language, so one could tell that specified callout should use a specified thread. Something like:

threadHandle := Smalltalk currentThreadId.

It is not hard to make these things accessible in the image, for example:

OSProcess thisOSProcess pthread ==> a ByteArray(240 166 184 224 255 255 255 255)

(this on a 64-bit platform, hence the 8-byte handle)

its not really matters what object represents the thread handle (it could be simple smallint). And surely, its not that hard to expose some thread-aware API to language level, given that VM _is_ thread-aware. But by not exposing it, we losing another lever of control, using which we could write a sophisticated software, purely in smalltalk without getting down at C level. That's what i ment to say.

Dave

Bert Freudenberg wrote:

On 10.11.2009, at 09:00, Andreas Raab wrote:

...

... make the Squeak VM be "truly event driven" that is invoke it in response to OS or other events instead of having the VM poll.

Don't have time to really get involved but I like the idea :)

I am interested as well. On a related issue, another programmer is working on the "announcements for i/o events" idea we briefly discussed in the beginning of the year.

About having the scheduler in the image rather than in the VM, that is how Self does it. A friend of mine started his PhD thesis using Squeak but switched to Self because of this feature even though I showed him that you could use a few tricks to change the behavior of the Squeak scheduler in interesting ways (I have implemented real-time scheduling with stuff like first-deadline-first, rate monotonic and so on using a higher priority thread waiting on the timer as the actual scheduler).

About to pros and cons of having platform dependent code in the image, there is an interesting discussion in the "Dorado" chapter of the green book. My proposal to the SqueakNOS group was that we could have two images: a small headless one with all the SqueakNOS drivers and primitives and a normal one with the user application. The second image would send messages to the first one for all its i/o so no code would directly access the hardware. In practice, you have the same protection as if there low level stuff were in the VM (or some underlying OS) and the image stays clean of any platform dependency. But if you really know what you are doing, you can call up remote debuggers, inspectors and browser on the first image and it will feel like all this stuff is implemented in the image instead of the VM.

-- Jecel

I believe that we now have a working configuration map for VMMaker and sundry plugins:

MCMcmUpdater updateFromRepositories: #('http://squeaksource.com/VMMaker' )

Once loaded, updates can be requested with VMMaker class>>updateFromServer.

So far, the MCM consists of Andreas recipe, plus a few of my favorites. We'll need to add references for other plugins needed for Mac OS, SqueakLand, etc.

Dave

On Tue, Nov 17, 2009 at 09:22:44PM -0800, Andreas Raab wrote:

Cool. Seems to work great. The two "unusual" things I noticed where MemoryAccess and SlangBrowser. How do I use those?

MemoryAccess is a version of the sqMemoryAccess.h macros implemented in Slang. To my surprise it seems to perform just as well as the C macros, and it is handy for inspecting the generated C code if you are looking for type declaration problems and improper casts. It is described here:

http://wiki.squeak.org/squeak/6081

SlangBrowser permits you to browse the generated C code interactively in the image. It is described here:

http://wiki.squeak.org/squeak/5916

There are some hacks required to get SlangBrowser integrated into the browsers (it's different for Squeak 3.8 and 3.9+). I have not yet tried getting this working for Squeak trunk, and I'll probably be needing some advice on that as I'm almost certainly not doing it in a good way.

Dave

Hi Ang -

I haven't completed the work. At this point I've been able to run the VM, execute a blocking FFI call (MessageBox, OpenFile) have the VM run happily in the background until I hit OK in the dialog. When the VM should return from the callback, it violently crashes :( There is some state that isn't preserved correctly but it's hard to find these things. I'll keep folks informed how this unfolds.

Cheers, - Andreas

Ang BeePeng wrote:

Hi,

I'd like to make my interpreter run until there's no more work to do, and return. I'm not good enough to put things that you explained into coding. Can you show or lead me to get it work?

Or, maybe, can you share the source of event-driven Squeak VM, so that I can learn from the code?

Thanks.

Ang Beepeng

Andreas Raab wrote:

...

having interpret() run until "there is no more work to do" and return from interpret() when it's all said and done. The trick is that instead of running the idle loop, the VM would determine that it has no more work to do when there is no runnable process, so when it finds that there is no runnable process it would return from interpret saying "my work's done here, there is no more code to run at this point, ask me again when an external event comes in".

The changes would be fairly straight forward: First, nuke the idle loop and allow wakeHighestPriority to return nil when there's no runnable process. Second, have transferTo: do a longjmp to the registered vmExitBuf to leave interpret(). Third, have interpret register the vmExitBuf and wake up the highest priorty process like here:

interpret "install jmpbuf for main interpreter" (self setjmp: vmExitBuf) == 0 ifTrue:[ self checkForInterrupts. "timers etc" "transferTo: longjmps if arg is nil so no need to check" self transferTo: self wakeHighestPriority.

     "this is the current interpret() implementation"
     self internalizeIPandSP.

self fetchNextBytecode. [true] whileTrue: [self dispatchOn: currentBytecode in: BytecodeTable].

 ].

At this point we can write a client loop that effectively looks like:

/* run the interpreter */ while(!done) { /* check for new events */ ioProcessEvents(); /* run processes resulting from the events */ interpret(); }

Now, obviously this is inefficient, we'd want to replace the ioProcessEvents() call with something more elaborate that reacts to the incoming OS events, takes the next scheduled delay into account, checks for socket handles etc. But I'm sure you're getting the idea. Instead of wasting our time in the idleProcess, we just return when there's no more work to do and it's up to the caller to run interpret() as often or as rarely as desired.

I also think that this scheme could be made backwards compatible by ensuring that we never call interpret() recursively. In this case an "old" image with the idle process would run the way it does today, and a "new" image without the idle process would live in the shiny new event driven world and return as needed.

What do you think? Any reasons why this wouldn't work?

Cheers,

• Andreas