C# 4 System.Threading.Tasks performance interrogations

I'm currently working on a ASP.NET MVC application with some pages loading a lot of data (repartited in separate LINQ queries).

To increase performance of these pages, i envisage to use the C# 4 Tasks to allow to make queries simultaneouly and, gain execution time.

But I have one major question : from a server side, wich situation is the best :

  • pages who use Tasks and so, a lot of the server resources in a small amount of time ?
  • pages who use only synchronous code, less server resource but a bigger amount of time ?
  • no incidence ?

Performance of my pages is important, but stability of the server is more ! Thank's by advance for your help.

Answers


You don't say whether the LINQ queries are CPU bound (e.g. computed in-memory) or IO bound (e.g. reading across the network or from disk).

If they are CPU bound, then using asynchronous code will improve fairness, but reduce throughput - but only so everyone suffers. For example, say you could only process one request at a time, and each request takes 5 seconds. Two requests come in almost at the same time. With synchronous code, the first will complete in 5 seconds while the second is queued, and the second will complete after 10. With asychronous code, both will start together and finish after slightly more than 10 seconds (due to overhead of swapping between the two). This is hypothetical, because you have many threads to process requests concurrently.

In reality, you'll find asynchronous code will only help when you have lots of IO bound operations that take long enough to cause request queuing. If the queue fills up, the server will start issuing Server Unavailable 503 errors. Check your performance counters - if you have little or no requests queued in ASP.NET under typical live loads then don't bother with the additional complexity.

If the work is IO bound, then using asynchronous code will push the bottleneck towards the network/disk. This is a good thing, because you aren't wasting your web server's memory resource on idle blocked request threads just waiting for responses - instead you make request throughput dependent on downstream performance and can focus on optimizing that. i.e. You'll keep those request threads free to take more work from the queue.

EDIT - Nice article on this here: http://blog.stevensanderson.com/2010/01/25/measuring-the-performance-of-asynchronous-controllers/


Need Your Help

stringbuffer toString does not match with same static string

php android http-post httpresponse stringbuffer

I am in middle of an issue where I am getting a string "INSERTED SUCCESSFULLY" as the response from my httprequest, my aim was to match it with a local string (same as above) however when I try to ...

How to mock Context using Mockito?

android bluetooth mockito android-wifi android-context

I'm using Context to access system level services like WifiManager and BluetoothManager. How to mock this getApplicationContext() using Mockito?