views:

479

answers:

6

I encountered a strange problem this week that I can't explain: I switched my application to use the signed version of some third party assemblies (Xceed Grid and some of their other components) and the application start time went into the toilet. Each time the application loaded a signed assembly, it took 30 seconds to load. Application start went from 5 seconds to over 90 seconds. What the heck is going on here?!

Some other info:

  • This is a WinForms app running under .NET 3.5 SP1.
  • The computer had no internet connection (on purpose, for security).
+1  A: 

Loading signed assemblies will definitely be slower than non-signed counterparts because signature needs to be verified, but this should be completely negligible.

Passing from 5 seconds to 90 seconds?? I think you need to contact the assembly author and ask them if they changed only the signature :-)

Darin Dimitrov
A: 

Maybe the signed assemblies are not NGEN'd, while the unsigned ones are.

leppie
Not sure but he is talking about 90!! seconds. Ngen is never going to have such a huge impact unless the assembly would be (literally) GBs in size.
Foxfire
+5  A: 

Hi there.

Have a look at these links:

http://blogs.technet.com/markrussinovich/archive/2009/05/26/3244913.aspx http://blogs.msdn.com/tess/archive/2008/05/13/asp-net-hang-authenticode-signed-assemblies.aspx

They might help. It could be that the config on your system means that the .NET framework is doing a lot of extra work to verify the assembly. If this is the case, then you can configure it to not be so picky.

Cheers. Jas.

Jason Evans
Also, contact the author of the assembly in case they have other customer's who have reported the same issue. They may have an FAW section on their site or something.
Jason Evans
Thanks, that was exactly the problem! Heading over to post on the Xceed forums now so no one else has to suffer this same pain.
Mike Post
A: 

I would guess that you have the security settings set in a way so that the assemblies certificates get verified. So it likely tries to access the web to verify some certificate and then waits for a timeout (30 sec is a VERY typical timeout number).

You can verify this if you look at what happens in that 30 seconds. For my guess to be true there should be little CPU use and little HDD accesses in those 90 seconds. If you have high CPU use or bound by your HDD then it's something else.

BTW: Another option would be if your HDD is completely full and the assemblies are EXTREMELY fragmented (but 90 secs would be more than I ever heard of in that case).

Foxfire
+1  A: 

Try to start you application from visual studio with "Step over". This will start the code by stepping over each app, so you can check what takes so long. I once had this, and it turned out that my sql server was really messed up.

Another way to find out why it takes so long is to place breakpoint scattered through the loading code and see what the bottleneck is. If the application takes 90 seconds before it your first like, probably something with XCeed, or loading the signed assemblies.

Btw, im aware there are better ways to profile your application, but this quick 'n dirty way works quite nice and efficient to debug such problems

Henri
+3  A: 

Jason Evans' post does contain the answer, but in the form of a link. I thought it would be good to post the actual solution here:

Create a file Appname.exe.config in the same folder as the executable (where Appname is the name of your executable; for development, this would be in the debug output folder). This shows an xml file that assumes you don't have other entries in the main config file; if you have the file already, I assume you would just add the new sections/text as necessary:

<?xml version="1.0" encoding="utf-8"?>
<configuration>
    <runtime>
        <generatePublisherEvidence enabled="false" />
    </runtime>
</configuration>

Martin

MartinKB
@unknown: Paste all your code into the post, highlight it, and click the "code" button on the formatting bar.
Will Eddins
Thanks Will.BTW, I see I should probably have posted my answer as a comment to Jason Evans' answer. I didn't mean to preempt his post. [Doh! What a newbie!]
MartinKB
Thanks for the answer! I still think the link is the definitive answer though because it contains the code, an explanation of the problem, and a thorough explanation of how to verify the problem.
Mike Post