I am building a background processing engine which supports discarding both to-be-processed and is-being-processed items. This is for usage in a winforms application that will require heavy processing of some input elements, so I m building a queue engine where I can enqueue workload items, and when they re processed, I get notified with the results.
The question is, this queue will almost always contain a lot of items to begin with, and I thought that instead of just dumping everything to the threadpool, I d place only the first N items into the threadpool, and keep backfilling when they are processed. The reason I want to do this is that once I dump them into the threadpool, they will be processed, and even if they re tagged as discard, they will still take up queue time.
With the backfill implementation I ve made, I can remove items from the queue if they become discarded, and only put them into the queue when it s their turn, so to speak.
So the question is, how would I go about calculating this number N, the number of items to place into and keep in the thread pool queue.
Issues I ve considered:
- I might want to enqueue 2 * number of processors, which I see is a typical number of items, to ensure all processors are working
- However, if the actual processing of some items is super-fast (which can happen), then the queue in the threadpool is exhausted before my own class can backfill with more work, so perhaps I d like a bigger number to avoid underutilizing the processors
- Should I create some auto-adjust routine to calculate the optimal number based on the current time each item takes, so that if they are all super-fast, the number is much higher, and if processing takes a bit of time, it should stay low?
What do you think?
New: Ok, due to one of the answers, I ll explain a bit more. Every item put into the queue is keyed by something unique. If I dump another item into the queue with the same key as an existing item, that old item is considered "Discarded", and should be removed. If the item is being processed, a property on the workload item is set to true, a "IsDicarded" property, which the processing method is responsible for calling. If it detects a discarded item, it should quit early, returning no results.
Perhaps I should experiment a bit more, and try to just dump everything into the threadpool.
New question: Is there a limit to the number of items I can queue up? If not, then this would easily simplify my class a lot.
Note: When I say "lengthy processing", I mean in the order of 1-10 seconds. Is the threadpool even the best for this? I see notes all over the web about "the processing should be quick", but what "quick" is is never mentioned. Is quick in the order of milliseconds here?