Async/Await vs Parellel.For, which is better in this instance?

Question

So I have 1000s of items to check whether they are up to date. Each one of those items requires reading thousands of files (some of which might be the same file across different items).

Currently this is implements using the TPL (async/await), one for each file it has to read and one for each item it has to check. This works fine, except for when I profile it, about the 3rd most expensive function is TrySteal in the thread pool.

Using the visual studio concurrency viewer, I see that 99% of a threads time in spent in concurrently related items, and only 1% in execution. It is this that leads me to think that I am perhaps just creating too many tasks (note: I don't use Task.Run anywhere, just await).

Would Parellel.For be any less overhead than reading a bunch of files using async/await? How much overhead is expected using the task programming library?

Answer 1

If you are checking files on the hard drive, I don't think that this task is very well parallelable. If you are trying to read thousands of files at the same time, you just make the process much slower, because it cannot read that many of them at the same time, and even worse, it cannot cache too many into memory. The fastest option, without optimization of the checking process itself, should be just running it consecutively. If you really want to optimize it, I suggest to loop through the files, checking for each item, instead of looping through item, checking each file. In this case, it might be effective even to do it in multiple threads (not all at once though).

Update: For the case when you have enough memory to cache all your files, then it does not restrict multithreading that much. Still, I would suggest to limit amount of parallel threads to number, comparable to amount of processor cores you going to work with. It is better to do it with Parallel.ForEach(). Also, Parallel.Foreach() clearly states, that you loop is async, so the code will be easier to understand.