Waiting on multiple Threads

  • Thread starter Thread starter MikeE
  • Start date Start date
M

MikeE

Hi all,

What's the best way to queue up and wait for number of threads to complete.
This problem was trivial in VC++ 6 but I'm finding it rather hard to solve
in VB.NET. My calculations run about 2 mins each (on a 2.8 Ghz Xeon and the
server range from 2 to 8 processors) (give or take about 15 secs). I have 8
sets of calculations to do (in another app I have 121) all the same calc
just different data. But the rest of the processing cannot continue until
ALL of the calculations are done.

so here's the algorithm I want to use (written from the C++ implentation
standpoint) {for ease of reading i have left out any/all discussion about
the queue ops used to schedule the runs which can change to what ever is
appropriate}
------------------------------------------------------------------
get user pref on number of concurrent threads...
for loop (to number of concurrent threads)
start a thread (with data)
add to wait list
while true
WaitForMultipleObjects(...threads...)
if more threads to process
launch thread with data
add to wait list
else if all threads finished
break
------------------------------------------------------------------

The problem I'm having is finding the .NET equivalent to
WaitForMultipleObjects. The threads might not complete in order or might
complete at the same time. I was toying with the idea of a delegate that
would basically have the functionality of the while loop, but seemed like
spaghetti coding to me.

Any ideas or examples you can think of would be greatly appreciated.

Thanks in advance,
Mike
 
What's the best way to queue up and wait for number of threads to complete.
This problem was trivial in VC++ 6 but I'm finding it rather hard to solve
in VB.NET. My calculations run about 2 mins each (on a 2.8 Ghz Xeon and the
server range from 2 to 8 processors) (give or take about 15 secs). I have 8
sets of calculations to do (in another app I have 121) all the same calc
just different data. But the rest of the processing cannot continue until
ALL of the calculations are done.
Click to expand...

That's easy then - just call Join() on all the threads.
 
Jon,

Doesn't Join() block until that particular thread is done (the same as
WaitForSingleObject)? If so, then that won't give me the proper scheduling
right? When any of the current threads end, I need to schedule the next in
the queue, and I can't schedule (i.e., start()) the next one UNTIL one of
the MAX # finishes.

Please correct me if i'm wrong about Join().

Thanks,
ME
 
Doesn't Join() block until that particular thread is done (the same as
WaitForSingleObject)?
Yes.

If so, then that won't give me the proper scheduling right?
Nope.

When any of the current threads end, I need to schedule the next in
the queue, and I can't schedule (i.e., start()) the next one UNTIL one of
the MAX # finishes.

Please correct me if i'm wrong about Join().
Click to expand...

Maybe I've misunderstood you. I thought you had created a bunch of
threads, started them running, and then wanted to know when they'd all
finished. In that case, calling Join() on each of them in turn does
exactly what you want.

If that's not what you're doing, you'll need to give a bit more
information about what you're doing.
 
Jon,

I'll try to restate then...

I have X number of tasks (unknown run time) that must be completed before
additional work is done. The USER specifies that the system can use N
threads to do the processing in parallel. How do we get this done in
VB.NET? The VC++ implentation is as follows...

1) Queue all work data (X items)
2) For i = 0; i < N; i++
2a) Get work data from queue (i.e., remove one item)
2b) Create Thread with work Data
2c) Start Thread
2d) Add Thread to watch list
3) while (true)
3a) WaitForMultipleObjects(watch list) <--- this will
block until ANY of the threads return
3b) remove finished thread from watch list
3c) If work data queue NOT empty
3c1) Get work data from queue (i.e., remove one item)
3c2) Create Thread with work Data
3c3) Start Thread
3c4) Add Thread to watch list
3d) else if all threads done (watch list empty)
3d1) break; <--
break the while loop when ALL data is processed and ALL threads are done
// else still processing threads...
4) additional work...


So another way to state my question is...
How do I block on multiple threads in VB.NET? Are WaitHandles my only
option?

Thanks,
 
Mike,
It sounds like you are wanting to write your own thread pool.

Any reason you are not using System.Threading.ThreadPool itself?


If ThreadPool was not quite what I needed, instead of creating a distinct
thread for each request I would simply create N threads. I would place all
the requests into a single System.Collections.Queue, each thread would read
a request from the queue, process it, then get another request from the
queue. Of course the Queue itself would need to be wrapped in thread safe
code such as via Queue.Synchronized. You may need to implement something to
let each thread know when it should stop processing requests...

Hope this helps
Jay

MikeE said:
Jon,

I'll try to restate then...

I have X number of tasks (unknown run time) that must be completed before
additional work is done. The USER specifies that the system can use N
threads to do the processing in parallel. How do we get this done in
VB.NET? The VC++ implentation is as follows...

1) Queue all work data (X items)
2) For i = 0; i < N; i++
2a) Get work data from queue (i.e., remove one item)
2b) Create Thread with work Data
2c) Start Thread
2d) Add Thread to watch list
3) while (true)
3a) WaitForMultipleObjects(watch list) <--- this will
block until ANY of the threads return
3b) remove finished thread from watch list
3c) If work data queue NOT empty
3c1) Get work data from queue (i.e., remove one item)
3c2) Create Thread with work Data
3c3) Start Thread
3c4) Add Thread to watch list
3d) else if all threads done (watch list empty)
3d1) break; <--
break the while loop when ALL data is processed and ALL threads are done
// else still processing threads...
4) additional work...


So another way to state my question is...
How do I block on multiple threads in VB.NET? Are WaitHandles my only
option?

Thanks,
Click to expand...
 
I'll try to restate then...

I have X number of tasks (unknown run time) that must be completed before
additional work is done. The USER specifies that the system can use N
threads to do the processing in parallel. How do we get this done in
VB.NET? The VC++ implentation is as follows...
Click to expand...

That sounds exactly like a thread pool. If the system thread pool
doesn't work for you (and there are various reasons why you might not
want to use it) you may want to look at the implementation I've got at
http://www.pobox.com/~skeet/csharp/miscutil

It's not been tested much, but you're welcome to it...
 
John/Jay,
Yes! what I want is so close to the threadpool that it is irritating. The
ONE MAJOR missing part for a thread pool is the MAX number of threads. See,
this processing is some rather lengthy numerical calculations, VERY CPU
intensive. We do not run it on any machine that does not have at least 2
processors (or one hyper threaded one, and then just for testing).
Otherwise it locks the system up. But we have some servers that have up to
8 CPUs in them. So we need the number of concurrent threads to be variable.
If we just lanched them all at once, we would QUICKLY run out of cpu
bandwith for the calcs and the system. (back to locked systems). we just
have to be able to control the number of "active" threads at one time.

What would your thoughts/feelings be towards using "WaitForMultipleObjects".
We of course could import it (which ive done) but having trouble setting up
the array of Thread handles to be watched. Any ideas? (just a reminder...i
kinda need this in vb.net)

On another note.. since i do have to get something working soon.. i've
started down the spaghetti code method, where the threads fire events to
which i have a re-enterable function that can access the remaining queue,
and when finally empty and done fires another event that is handled by a
delegate that does "the rest of the stuff" Unfortunatly this is looking
REALLY ugly, so i'm trying to come up with a class to wrap the thread and
that contains a wait handle which is montitored.. but still trying to flush
this out.

Again, what gets me, is that this takes less than 50 lines of code in VC++,
with error checking, and was rock solid. I'm really having a tough time
coming to grips with the fact that we can't do it EVEN EASIER in .NET like
so many of our other upgrades.

And john, thanks for the link.. i'll take a look at miscutil tomorrow!

Thanks again for your time on this,
ME
 
Yes! what I want is so close to the threadpool that it is irritating. The
ONE MAJOR missing part for a thread pool is the MAX number of threads. See,
this processing is some rather lengthy numerical calculations, VERY CPU
intensive. We do not run it on any machine that does not have at least 2
processors (or one hyper threaded one, and then just for testing).
Otherwise it locks the system up. But we have some servers that have up to
8 CPUs in them. So we need the number of concurrent threads to be variable.
If we just lanched them all at once, we would QUICKLY run out of cpu
bandwith for the calcs and the system. (back to locked systems). we just
have to be able to control the number of "active" threads at one time.

What would your thoughts/feelings be towards using "WaitForMultipleObjects".
We of course could import it (which ive done) but having trouble setting up
the array of Thread handles to be watched. Any ideas? (just a reminder...i
kinda need this in vb.net)
Click to expand...

Well, you don't really want to wait for other threads to actually
finish, because you can't reuse them. Doing a thread pool of *some*
description (which could just be several threads with a
producer/consumer queue - see
http://www.pobox.com/~skeet/csharp/threads/deadlocks.shtml for some
sample code) is a much better answer here.

Alternatively, using Auto/ManualResetEvents, one per thread, would
allow you to wait for any of the threads to finish without being too
hard. Or just have a single monitor which the main thread waits on and
which other threads pulse.
On another note.. since i do have to get something working soon.. i've
started down the spaghetti code method, where the threads fire events to
which i have a re-enterable function that can access the remaining queue,
and when finally empty and done fires another event that is handled by a
delegate that does "the rest of the stuff" Unfortunatly this is looking
REALLY ugly, so i'm trying to come up with a class to wrap the thread and
that contains a wait handle which is montitored.. but still trying to flush
this out.

Again, what gets me, is that this takes less than 50 lines of code in VC++,
with error checking, and was rock solid. I'm really having a tough time
coming to grips with the fact that we can't do it EVEN EASIER in .NET like
so many of our other upgrades.
Click to expand...

You may well find it's easier in .NET, once you change the approach
slightly. There are lots of things which are hard to do if you try
using the "old" approach, but for which there are new approaches which
make things easier.
 
Jon,

Thanks, I'll look into the single monitor option... seems promising.

One clarification though.. I HAVE to wait for all the threads to finish, not
because I want to reuse the thread, but because I need the "ANSWER". A math
problem is not finished until all the parts are there. In our program the
final result can NOT be calculated until ALL the threads have returned.

Thanks again,
ME
 
MikeE,
Why not use two queues?

One queue for the questions (requests) to the N threads in the pool and a
second queue for the answers (to the main thread). The main thread can
simply wait for answers on the second queue, once it has all the answers it
can figure the final result.

Hope this helps
Jay

MikeE said:
Jon,

Thanks, I'll look into the single monitor option... seems promising.

One clarification though.. I HAVE to wait for all the threads to finish,
not because I want to reuse the thread, but because I need the "ANSWER". A
math problem is not finished until all the parts are there. In our program
the final result can NOT be calculated until ALL the threads have
returned.

Thanks again,
ME
Click to expand...
 
Back
Top