views:

94

answers:

2

I just happen to check the performance of an ASP.NET MVC application we are building. I was going to insert a partial view into a loop, and just out of curiosity I checked how long it took to render the page. The result was not good.

I need to do more conclusive investigation, but just in case there was somebody with similar issues or more insight, here is what I have so far. First, I should say that all results and measurements were done after multiple page loads and that I have set <compilation debug="false"> in my web.config.

  • It seems that a single render partial incurs about 5ms hit (at least in my environment). When I inline the actual content of the partial view, I get practically 0ms.
  • When I include an empty partial view to a loop of about 70 elements, the total render time increases by ~ 60ms. So there is some caching presumably, but it's not ideal.
  • I debugged ASP.NET MVC, and found out that partial views are cached, but it only caches the paths to the ascx's. The actual views are then instantiated every time using the BuildManager.CreateInstanceFromVirtualPath method.
  • And now the interesting bit: When include the same partial view using the WebForms syntax (<my:UserContol runat="server" />), the extra 60ms go away.

So based on the observations above, it seems the culprit is the BuildManager.CreateInstanceFromVirtualPath method. Maybe, it was not meant to be called multiple times. Webforms presumably don't use it; or use it somehow only once for each ascx?

A: 

I am guessing the answer is ... it depends?

Partial views decrease performance (the overhead of the actual call etc).

Partial views are not cached.

Including a partial view inside a loop will decrease performance, and can be sped up again by moving the loop inside the partialview instead.

Some sample reading (which references the caching of the viewpath) can be found here.

aanund
The real performance hog is *recursive partial views*. While certainly a fitting concept in some cases (a tree menu comes to mind), performance unexpectedly degraded worse than linearly when I last tried it.
bzlm
A: 

60ms is such a small interval it sounds like statistical noise to me, not conclusive evidence of a performance difference.

Haacked