我应该在一个类中封装Task吗？

Question

I'm building an application having multiple long running worker threads which need to be running 24x7. Each worker is a method in a class called Bot . Start/Stop of the Bot is controlled by the class itself. So if I need 10 bots running, I just need to instantiate 10 Bot , store in a List<Bot> , and start everyone of them. I also have a BotManager class to manage all the currently running Bot

Now, I wanted to use Task instead of Thread for each of the workers, as they're only spending less than 5% of the time doing any processing.

Should I keep a Task in each class or should I inherit my Bot from Task ?

Answer 1

I wouldn't descend from Task . You're not really extending the idea of "task" and making something that, to code that uses it, is a task with extra features; instead you're just making something that happens to use tasks for its implementation.

I'm also not sure Task is a good fit here. If the idea is that these keep running in the background, then that's not playing to TPL's strengths. The best time to use Task is when you need composability -- a well-defined task that runs, completes, and then notifies other tasks that it's done and they can start working on its results.

Answer 2

As I understood, your requirement is to run as many thread as possible to optimize the CPU usages. And the same time you want framework to take care of this responsibility to increase and decrease the current running threads according to CPU usage.

I would say TPL (Tasks) would the better option. Please look into http://msdn.microsoft.com/en-us/library/dd537609.aspx

We can help you more if you share the nature of your work you want to perform in parallel. What do you exactly mean by 24X7 running tasks? What kind of work your tasks perform? Is it computing or IO (DB operation/Web Service Call)?

I am assuming that each request consists of 4 steps and these steps are executed in following sequence, 1. DB Look-up 2. Web service call 3. DB Update 4. Log to File

As you mentioned Web service call takes time and therefore best suited for asynchronous I/O task.

Next you said,

CPU intensive part will be the db lookup

But I understand that db lookup does not use CPU and it is an I/O activity and again good candidate for asynchronous I/O task.

Considering all these assumptions I suggest following,

1. DB Look-up (Run asynchronously so that thread pool thread will be freed to process other requests)
2. Web service call (Run asynchronously so that thread pool thread will be freed to process other requests)
3. DB Update (Run synchronously)
4. Log to File (use asynchronous logging depending upon log content size)

According to my understanding none of your steps are CPU intensive that is why you are seeing only 5% of CPU usages.

You also mentioned,

followed by a random delay to distribute the load evenly.

Did you mean to run as much as possible such requests (consists of these 4 steps)? If so then you do not need to put any extra stuff. Because when a request waits for asynchronous call then it frees the thread pool thread and system can use it to process other request. This will increase the throughput.

Should I keep a Task in each class or should I inherit my Bot from Task?

1 Bot represent to 1 request. Tasks represent to Steps within a request. So create tasks within your BOT class.

Run it with different number of requests and verify the usages of CPU, Memory and other system resources.

Answer 3

This is definitely not a "task" for tasks. Use threads. When you need assurance, that your code is actually running (and on a fixed amount of threads) you cannot use tasks . Tasks do not guarantee parallel execution. You must allocate threads yourself. If you want to save resources reduce the threads stack size using the constructor.