Performance Quiz #6 -- Looking at the fourth cut

.NET Efficiency

Alan's Corner — Tue, 31 May 2005 16:26:14 GMT

So I was reading through one of my favorite MSDN blogs (http://blogs.msdn.com/oldnewthing/)

And he...

re: Performance Quiz #6 -- Looking at the fourth cut

stefang — Tue, 17 May 2005 18:27:11 GMT

Rico, do you have any comments to my comment at:

http://blogs.msdn.com/ricom/archive/2005/05/12/416977.aspx#418471

re: Performance Quiz #6 -- Looking at the fourth cut

James Clarke — Tue, 17 May 2005 18:22:52 GMT

Rico.. Do you have any working set numbers to compare between the two implementations?

re: Performance Quiz #6 -- Looking at the fourth cut

Will Dean — Tue, 17 May 2005 12:53:31 GMT

The managed version takes around 50ms to run here, built with VS2005B2. (My PC is faster than I thought...)

About 53% of the time (including children) is spent in Parse, about 23% in Substring in Parse.

I don't know if the VS profiler can be made to tell me more about what's happening INSIDE something like Substring.

My approach (as a C programmer, really) my be to load the whole string and just use parse to calculate the offsets and lengths of everything, thereby avoiding all the string copying.

But of course, that would just be a trade-off of load time vs. access time, and I don't know enough about the rest of the application to know if that would be a good thing.

re: Performance Quiz #6 -- Looking at the fourth cut

ricom — Tue, 17 May 2005 11:25:45 GMT

I did see Raymond's program quite some time ago, but I didn't do any analysis then. Still it's not hard to predict where he's likely to go because he keeps picking off the biggest performance problem and it's easy enough to spot that by looking at what he's done so far. It's not like I'm doing some exotic analysis here :)

re: Performance Quiz #6 -- Looking at the fourth cut

Asd — Tue, 17 May 2005 10:30:32 GMT

Just out of curiosity I did a direct port of the managed version to Java. I'm getting around 203ms.

re: Performance Quiz #6 -- Looking at the fourth cut

Larry Osterman [MSFT] — Tue, 17 May 2005 08:05:54 GMT

I've been trying to figure if you and Raymond are doing this in concert - especially since you seem to be giving the hints to where Raymond's going to be going next...

And I'm still waiting for him to totally rip the CRT out of the picture.

He's pretty close right now. All he has to do is get rid of those darned string thingies (which are fundimentally pointless since he's dealing with what are effectively fixed length items).

And no, I didn't peek at his next posts.

re: Performance Quiz #6 -- Looking at the fourth cut

Dean Harding — Tue, 17 May 2005 02:07:34 GMT

Hehe, you going to hold your trump card in reserve until the unmanaged one is as fast as it'll get? :)

re: Performance Quiz #6 -- Looking at the fourth cut

ricom — Mon, 16 May 2005 23:52:46 GMT

I'm saving the managed analysis for a bit... partly waiting to see if any of my readers have more to say based on their own measurements and partly waiting for Raymond's unmanaged to finally beat the managed time.

I think it will be interesting to see what it takes and what finally did the trick. Hang in there a few more days.

I think the whole exercise is just full of surprises so I'm kinda hoping more people will say what's on their mind.

re: Performance Quiz #6 -- Looking at the fourth cut

Michael S. Kaplan — Mon, 16 May 2005 23:34:00 GMT

Hey Rico!

I'm still aching to know how the Everett vs. Whidbey numbers look. :-)