Cold Startup Performance

re: Cold Startup Performance

Matt — Tue, 19 Oct 2004 01:20:00 GMT

Wouldn't registry access be fairly low cost (at least in comparison to config files) given it is widely used by the OS and other processes? Surely a substantial part of the registry would be cached by the time any managed code is started?

re: Cold Startup Performance

Rico Mariani — Tue, 19 Oct 2004 01:30:00 GMT

Maybe, maybe not. We are talking about cold startup after all so presumably not too much has run. And the registry is not created equal -- the portions you need may or may not be the same as what the operating system already has accessed.

re: Cold Startup Performance

Saurabh Jain — Tue, 19 Oct 2004 10:25:00 GMT

Does the sharing on ngen'd images happen across concurrent login sessions? Consider the case where a school has a server on which various students logon during a lab. Does it make sense to ngen everything to get maximum sharing and reduce memory use?

re: Cold Startup Performance

Rico Mariani — Tue, 19 Oct 2004 16:57:00 GMT

The native image cache is machine-wide so yes multiple users can share pages from the same images.

re: Cold Startup Performance

Rico Mariani — Tue, 19 Oct 2004 21:45:00 GMT

Updated the main body with some data from a colleague.

re: Cold Startup Performance

Saurabh Jain — Wed, 20 Oct 2004 09:02:00 GMT

Thanks Rico. The follow up question is, does the sharing of ngen images happen if the same dll (strongly signed) is installed at two different locations.

re: Cold Startup Performance

Frank Hileman — Wed, 20 Oct 2004 21:37:00 GMT

The data you give is not encouraging. How big were these dlls being loaded? It is proportional to size, generally speaking? What is the rough delta for a 350K assembly?

Thanks.

re: Cold Startup Performance

Rico Mariani — Wed, 20 Oct 2004 22:09:00 GMT

Hmmm... I was going to say "no" off hand but then it hit me that it's more subtle than that. If the strong name is the same then you'd expect the native image to land in the same place in the native image cache -- it wouldn't matter which flavor of IL you tried to load, both would use the same native bits so there would be sharing. So I double checked that last assumption and sure enough that is in fact the case -- the same native bits are used even if you get the (same) IL from different locations.

re: Cold Startup Performance

Rico Mariani — Wed, 20 Oct 2004 22:12:00 GMT

Frank: The cost isn't going to be proportional to size necessarily although size will be a leading indicator. It's a function of how much of the DLL you touch at startup and how you touch it. If you dance all over the place looking at a function here and a function there then you will drag in a lot of pages, most of which will be wasted because you used only a small fraction. If on the other hand you are careful about which methods you touch during startup then you will find that you can use a much smaller slice -- it's more about density than it is about size.

This isn't really a new problem, cold startup has been touchy like this since as long as I've been programming. We used to do swap tuning of linker overlays in the bad old days, it was really the exact same issue.

NGEN and Cold PC Starts

Mike Taulty's Weblog — Thu, 21 Oct 2004 17:50:00 GMT

re: Cold Startup Performance

Frank Hileman — Fri, 22 Oct 2004 20:24:00 GMT

Thanks. I asked because the numbers you give are so large -- I did not understand how it could be so slow, unless the dlls are huge. 350K is the size of a dll we produce, which is dependent only on System, System.Windows.Forms, and System.Drawing. I am not sure of the cold start, but the warm start seems instantaneous with our test apps. We cannot test without the ngen on the MS libs.

re: Cold Startup Performance

Junfeng Zhang — Wed, 27 Oct 2004 05:14:00 GMT

Saurabh said:

"The follow up question is, does the sharing of ngen images happen if the same dll (strongly signed) is installed at two different locations."

A better question will be: Why do you want to put the same dll in different locations? Why don't you put it in GAC?

Cold Startup Performance

dotRob — Fri, 29 Oct 2004 15:47:00 GMT

Cold Startup Performance

dotRob — Fri, 29 Oct 2004 15:54:00 GMT

NGEN and Startup Performance

Junfeng Zhang — Tue, 02 Nov 2004 10:44:00 GMT

Post #2

VIOLATING THE CODE — Mon, 27 Dec 2004 21:14:00 GMT

Post #2

VIOLATING THE CODE — Mon, 27 Dec 2004 21:16:00 GMT

Post #2

VIOLATING THE CODE — Tue, 28 Dec 2004 08:48:00 GMT

RSSBandit Performance Problems

notgartner.com: Mitch Denny's Blog — Tue, 09 Aug 2005 14:42:00 GMT

RSSBandit Performance Problems

notgartner.com: Mitch Denny's Blog — Tue, 09 Aug 2005 14:43:32 GMT

Programming Using .NET Can Be Such a Pain Sometimes...

Dare Obasanjo aka Carnage4Life — Sat, 13 Aug 2005 06:52:13 GMT

Project anti-pattern: Many projects in a Visual Studio Solution File

Chad Myers' Blog — Wed, 16 Jul 2008 04:18:05 GMT

I've been hearing from several colleagues about how their Visual Studio solution files have many

First vn.net app to launch takes ages | keyongtech

First vn.net app to launch takes ages | keyongtech — Thu, 22 Jan 2009 10:34:15 GMT

PingBack from http://www.keyongtech.com/478621-first-vn-net-app-to