I have a fairly long-running parfor loop (let's say 100,000 iterations where each iterations takes about a minute) that I'm running with 36 cores. I've noticed that near the end of the job, a large number of the cores go idle while a few finish what I think has to be multiple iterations per worker. This leads to a lot of wasted computing time waiting for one worker to finish several jobs while the others are sitting idle.
The following script shows the issue (using the File Exchange utility Par.m):
% Set up parallel pool
nLoop = 480;
p1 = gcp;
% Run a loop
pclock = Par(nLoop);
parfor iLoop = 1:nLoop
Par.tic;
pause(0.1);
pclock(iLoop) = Par.toc;
end
stop(pclock);
plot(pclock);
% Process the timing info:
runs = [[pclock.Worker]' [pclock.ItStart]' [pclock.ItStop]'];
nRuns = arrayfun(#(x) sum(runs(:,1) == x), 1:max(runs));
starts = nan(max(nRuns), p1.NumWorkers);
ends = nan(max(nRuns), p1.NumWorkers);
for iS = 1:p1.NumWorkers
starts(1:nRuns(iS), iS) = sort(runs(runs(:, 1) == iS, 2));
ends(1:nRuns(iS), iS) = sort(runs(runs(:, 1) == iS, 3));
end
firstWorkerStops = min(max(ends));
badRuns = starts > firstWorkerStops;
nBadRuns = sum(sum(badRuns)) - (p1.NumWorkers-1);
fprintf('At least %d (%3.1f%%) iterations run inefficiently.\n', ...
nBadRuns, nBadRuns/nLoop * 100);
The way I'm looking at it, every worker should be busy until the queue is empty, after which all workers sit idle. But here it looks like that's not happening - with 480 iterations, I'm getting between 6-20 iterations that start on a worker after a different worker has been sitting idle for a full cycle. This number appears to scale linearly with the number of loop iterations coming in near 2% of the total. With limited testing this appears to be consistent across Matlab 2016b and 2014b.
Is there any reason that this is the expected behavior or is this simply a poorly written scheduler in the parfor implementation? If so, how can I structure this so I'm not sitting around with idle workers so long?
I think this explains what you are observing.
If there are more iterations than workers, some workers perform more than one loop iteration; in this case, a worker might receive multiple iterations at once to reduce communication time. (From "When to Use parfor")
Towards the end of a loop, a two workers may finish their iterations at around the same time. If there is only one group of iterations left to be assigned, then one worker will get them all, and the other will remain idle. It sounds like it is expected behavior, and it is probably because the underlying implementation tries to reduce the communication cost associated with a worker pool. I've looked around the web and the Matlab settings and it doesn't seem like there is a way to adjust the communication strategy.
The parfor scheduler attempts to load-balance for loops where the iterations do not take a uniform amount of time. Unfortunately, as you observe, this can lead to workers becoming idle at the end of the loop. With parfor, you don't have control over the division of work; but you could use parfeval to divide your work up into even chunks - that might give you better utilisation. Or, you could even use spmd in conjunction with a for-drange loop.
Related
I am writing a matlab code, which does some operations on a large matrix. First I create three 3D array
dw2 = 0.001;
W2 = [0:dw2:1];
dp = 0.001;
P1 = [dp:dp:1];
dI = 0.001;
I = [1:-dI:0];
[II,p1,ww2] = ndgrid(I,P1,W2);
Then my code basically does the following
G = [0:0.1:10]
Y = zeros(length(G),1)
for i = 1:1:length(G)
g = G(i);
Y(i) = myfunction(II,p1,ww2,g)
end
This code roughly takes about 100s, with each iteration being nearly 10s.
However, after I start parfor
ProcessPool with properties:
Connected: true
NumWorkers: 48
Cluster: local
AttachedFiles: {}
AutoAddClientPath: true
IdleTimeout: 30 minutes (30 minutes remaining)
SpmdEnabled: true
Then it is like running forever. The maximum number of workers is 48. I've also tried 2, 5, 10. All of these are slower than non-parallel computing. Is that because matlab copied II,p1,ww2 48 times and that causes the problem? Also myfunction involves a lot of vectorization. I have already optimized the myfunction. Will that lead to slow performance of parfor? Is there a way to utilize (some of) the 48 workers to speed up the code? Any comments are highly appreciated. I need to run millions of cases. So I really hope that I can utilize the 48 workers in some way.
It seems that you have large data, and a lot of cores. It is likely that you simply run out of memory, which is why things get so slow.
I would suggest that you set up your workers to be threads, not separate processes.
You can do this with parpool('threads'). Your code must conform to some limitations, not all code can be run this way, see here.
In thread-based parallelism, you have shared memory (arrays are not copied). In process-based parallelism, you have 48 copies of MATLAB running on your computer at the same time, each needing their own copy of your data. That latter system was originally designed to work on a compute cluster, and was later retrofitted to work on a single machine with two or four cores. I don’t think it was ever meant for 48 cores.
If you cannot use threads with your code, configured your parallel pool to have fewer workers. For example parpool('local',8).
For more information, see this documentation page.
I have implemented a combinatorial search algorithm (for comparison to a more efficient optimization technique) and tried to improve its runtime with parfor.
Unfortunately, the work assignments appear to be very badly unbalanced.
Each subitem i has complexity of approximately nCr(N - i, 3). As you can see, the tasks i < N/4 involve significantly more work than i > 3*N/4, yet it seems MATLAB is assigning all of i < N/4 to a single worker.
Is it true that MATLAB divides the work based on equally sized subsets of the loop range?
No, this question cites the documentation saying it does not.
Is there a convenient way to rebalance this without hardcoding the number of workers (e.g. if I require exactly 4 workers in the pool, then I could swap the two lowest bits of i with two higher bits in order to ensure each worker received some mix of easy and hard tasks)?
I don't think a full "work-stealing" implementation is necessary, perhaps just assigning 1, 2, 3, 4 to the workers, then when 4 completes first, its worker begins on item 5, and so on. The size of each item is sufficiently larger than the number of iterations that I'm not too worried about the increased communication overhead.
If the loop iterations are indeed distributed ahead of time (which would mean that in the end, there is a single worker that will have to complete several iterations while the other workers are idle - is this really the case?), the easiest way to ensure a mix is to randomly permute the loop iterations:
permutedIterations = randperm(nIterations);
permutedResults = cell(nIterations,1); %# or whatever else you use for storing results
%# run the parfor loop, completing iterations in permuted order
parfor iIter = 1:nIterations
permutedResults(iIter) = f(permutedIterations(iIter));
end
%# reorder results for easier subsequent analysis
results = permutedResults(permutedIterations);
the code I'm dealing with has loops like the following:
bistar = zeros(numdims,numcases);
parfor hh=1:nt
bistar = bistar + A(:,:,hh)*data(:,:,hh+1)' ;
end
for small nt (10).
After timing it, it is actually 100 times slower than using the regular loop!!! I know that parfor can do parallel sums, so I'm not sure why this isn't working.
I run
matlabpool
with the out-of-the-box configurations before running my code.
I'm relatively new to matlab, and just started to use the parallel features, so please don't assume that I'm am not doing something stupid.
Thanks!
PS: I'm running the code on a quad core so I would expect to see some improvements.
Making the partitioning and grouping the results (overhead in dividing the work and gathering results from the several threads/cores) is high for small values of nt. This is normal, you would not partition data for easy tasks that can be performed quickly in a simple loop.
Always perform something challenging inside the loop that is worth the partitioning overhead. Here is a nice introduction to parallel programming.
The threads come from a thread pool so the overhead of creating the threads should not be there. But in order to create the partial results n matrices from the bistar size must be created, all the partial results computed and then all these partial results have to be added (recombining). In a straight loop, this is with a high probability done in-place, no allocations take place.
The complete statement in the help (thanks for your link hereunder) is:
If the time to compute f, g, and h is
large, parfor will be significantly
faster than the corresponding for
statement, even if n is relatively
small.
So you see they mean exactly the same as what I mean, the overhead for small n values is only worth the effort if what you do in the loop is complex/time consuming enough.
Parforcomes with a bit of overhead. Thus, if nt is really small, and if the computation in the loop is done very quickly (like an addition), the parfor solution is slower. Furthermore, if you run parforon a quad-core, speed gain will be close to linear for 1-3 cores, but less if you use 4 cores, since the last core also needs to run system processes.
For example, if parfor comes with 100ms of overhead, and the computation in the loop takes 5ms, and if we assume that speed gain is linear up to 4 cores with a coefficient of 1 (i.e. using 4 cores makes the computation 4 times faster), nt needs to be about 30 for you to achieve a speed gain with parfor (150ms with for, 132ms with parfor). If you were to run only 10 iterations, parfor would be slower (50ms with for, 112ms with parfor).
You can calculate the overhead on your machine by comparing execution time with 1 worker vs 0 workers, and you can estimate speed gain by making a liner fit through the execution times with 1 to 4 workers. Then you'll know when it's useful to use parfor.
Besides the bad performance because of the communication overhead (see other answers), there is another reason not to use parfor in this case. Everything which is done within the parfor in this case uses built-in multithreading. Assuming all workers are running on the same PC there is no advantage because a single call already uses all cores of your processor.
the code I'm dealing with has loops like the following:
bistar = zeros(numdims,numcases);
parfor hh=1:nt
bistar = bistar + A(:,:,hh)*data(:,:,hh+1)' ;
end
for small nt (10).
After timing it, it is actually 100 times slower than using the regular loop!!! I know that parfor can do parallel sums, so I'm not sure why this isn't working.
I run
matlabpool
with the out-of-the-box configurations before running my code.
I'm relatively new to matlab, and just started to use the parallel features, so please don't assume that I'm am not doing something stupid.
Thanks!
PS: I'm running the code on a quad core so I would expect to see some improvements.
Making the partitioning and grouping the results (overhead in dividing the work and gathering results from the several threads/cores) is high for small values of nt. This is normal, you would not partition data for easy tasks that can be performed quickly in a simple loop.
Always perform something challenging inside the loop that is worth the partitioning overhead. Here is a nice introduction to parallel programming.
The threads come from a thread pool so the overhead of creating the threads should not be there. But in order to create the partial results n matrices from the bistar size must be created, all the partial results computed and then all these partial results have to be added (recombining). In a straight loop, this is with a high probability done in-place, no allocations take place.
The complete statement in the help (thanks for your link hereunder) is:
If the time to compute f, g, and h is
large, parfor will be significantly
faster than the corresponding for
statement, even if n is relatively
small.
So you see they mean exactly the same as what I mean, the overhead for small n values is only worth the effort if what you do in the loop is complex/time consuming enough.
Parforcomes with a bit of overhead. Thus, if nt is really small, and if the computation in the loop is done very quickly (like an addition), the parfor solution is slower. Furthermore, if you run parforon a quad-core, speed gain will be close to linear for 1-3 cores, but less if you use 4 cores, since the last core also needs to run system processes.
For example, if parfor comes with 100ms of overhead, and the computation in the loop takes 5ms, and if we assume that speed gain is linear up to 4 cores with a coefficient of 1 (i.e. using 4 cores makes the computation 4 times faster), nt needs to be about 30 for you to achieve a speed gain with parfor (150ms with for, 132ms with parfor). If you were to run only 10 iterations, parfor would be slower (50ms with for, 112ms with parfor).
You can calculate the overhead on your machine by comparing execution time with 1 worker vs 0 workers, and you can estimate speed gain by making a liner fit through the execution times with 1 to 4 workers. Then you'll know when it's useful to use parfor.
Besides the bad performance because of the communication overhead (see other answers), there is another reason not to use parfor in this case. Everything which is done within the parfor in this case uses built-in multithreading. Assuming all workers are running on the same PC there is no advantage because a single call already uses all cores of your processor.
I'm working with a long running parfor loop in matlab.
parfor iter=1:1000
chunk_of_work(iter);
end
There are generally about 2-3 timing outliers per run. That is to say for every 1000 chunks of work performed there are 2-3 that take about 100 times longer than the rest. As the loop nears completion, the workers that evaluated the outliers continue to run while the rest of the workers have no computational load.
This is consistent with the parfor loop distributing work statically. This is in contrast with the documentation for the parallel computing toolbox found here:
"Work distribution is dynamic. Instead of being allocated a fixed
iteration range, the workers are allocated a new iteration only after
they finish processing their current iteration, which results in an
even work load distribution."
Any ideas about what's going on?
I think the doc you quote has a pretty good description what is considered a static allocation of work: each worker "being allocated a fixed iteration range". For 4 workers, this would mean the first being assigned iter 1:250, the second iter 251:500,... or the 1:4:100 for the first, 2:4:1000 for the second and so on.
You did not say exactly what you observe, but what you describe is well consistent with dynamic workload distribution: First, the four (example) workers work on one iter each, the first one that is finished works on a fifth, the next one that is done (which may well be the same if three of the first four take somewhat longer) works on a sixth, and so on. Now if your outliers are number 20, 850 and 900 in the order MATLAB chooses to process the loop iterations and each take 100 times as long, this only means that the 21st to 320th iterations will be solved by three of the four workers while one is busy with the 20th (by 320 it will be done, now assuming roughly even distribution of non-outlier calculation time). The worker being assigned the 850th iteration will, however, continue to run even after another has solved #1000, and the same for #900. In fact, if there were about 1100 iterations, the one working on #900 should be finished roughly at the time when the others are.
[edited as the orginal wording implied MATLAB would still assign the iterations of the parfor loop in order from 1 to 1000, which should not be assumed]
So long story short, unless you find a way to process your outliers first (which of course requires you to know a priori which ones are the outliers, and to find a way to make MATLAB start the parfor loop processing with these), dynamic workload distribution alone cannot avoid the effect you observe.
Addition: I think, however, that your observation that as "the loop nears completion, the worker*s* that evaluated the outliers continue to run" seems to imply at least one of the following
The outliers somehow are among the last iterations MATLAB starts to process
You have many workers, in the order of magnitude of the number of iterations
Your estimate of the number of outliers (2-3) or your estimate of their computation time penalty (factor 100) is too low
The work distribution in PARFOR is somewhat deterministic. You can observe precisely what's going on by having each worker log to disk how things go, but basically it turns out that PARFOR divides your loop up into chunks in a deterministic way, but farms them out dynamically. Unfortunately, there's currently no way to control that chunking.
However, if you cannot predict which of your 1000 cases are going to be outliers, it's hard to imagine an efficient scheme for distributing the work.
If you can predict your outliers, you might be able to take advantage of the fact that roughly speaking, PARFOR executes loop iterations in reverse order, so you could put them at the "end" of the loop so work starts on them immediately.
The problem you face is well described in #arne.b's answer, I have nothing to add to that.
But, the parallel compute toolbox does contain functions for decomposing a job into tasks for independent execution. From your question it's not possible to conclude either that this is suitable or that this is not suitable for your application. If it is, the general strategy is to break the job into tasks of some size and have each processor tackle a task, when finished go back to the stack of unfinished tasks and start on another.
You might be able to decompose your problem such that one task replaces one loop iteration (lots of tasks, lots of overhead in managing the computation but best load-balancing) or so that one task replaces N loop iterations (fewer tasks, less overhead, poorer load-balancing). Jobs and tasks are a little trickier to implement than parfor too.
As an alternative to PARFOR, in R2013b and later, you can use PARFEVAL and divide up the work any way you see fit. You could even cancel the 'timing outliers' once you've got sufficient results, if that's appropriate. There is, of course, overhead when dividing up your existing loop into 1000 individual remote PARFEVAL calls. Perhaps that's a problem, perhaps not. Here's the sort of thing I'm imagining:
for idx = 1:1000
futures(idx) = parfeval(#chunk_of_work, 1, idx);
end
done = false; numComplete = 0;
timer = tic();
while ~done
[idx, result] = fetchNext(futures, 10); % wait up to 10 seconds
if ~isempty(idx)
numComplete = numComplete + 1;
% stash result
end
done = (numComplete == 1000) || (toc(timer) > 100);
end
% cancel outstanding work, has no effect on completed futures
cancel(futures);