[Mono-dev] Performance of compiled regular expressions?

[Jordan Earls]

I actually found out how to use the profiler a bit better and found
that MarkdownSharp(http://code.google.com/p/markdownsharp/) is what
makes extensive use of regular expressions, however, the majority of
it's regular expressions are also compiled. (though are of course
significantly more complex).

Now, on to your questions

On Sun, Dec 2, 2012 at 10:03 PM, Daniel Lo Nigro <lists at dan.cx> wrote:
> Are you sure your site is running in release mode (debug="false" in your
> web.config, and Release build config for assemblies)? Running in debug mode
> will turn off lots of optimisations that may cause slowdown.

Yes, it's in release mode with debug=false

> It might be worth trying out Regex.CompileToAssembly to precompile your
> regular expressions - This will create a separate .NET assembly which has
> pre-optimised versions of the regular expressions. However I've found this
> is only beneficial for very large regular expressions, and regular
> expressions that are used a lot. So it might not help too much (but it
> doesn't hurt to try it to see if it makes a difference or not). As far as I
> know, Mono doesn't support Regex.CompileToAssembly, but you should be able
> to generate the assembly on Microsoft .NET, and then use the generated
> assembly with Mono. I haven't tried this myself though.

That's an option, but I'd really prefer to avoid it. My regular
expressions are easily cacheable and used quite a bit, but not complex
at all.

> Out of curiosity, how are you profiling the code?

I'm using this:

mono --profile=log:noalloc,calls,maxframes=16 /usr/lib/mono/4.0/xsp4.exe

I know xsp isn't a production server or anything, but I don't think it
matters much in this case. (I see the same kind of speed problems with
mod_mono and apache).

I then read the report like this:

mprof-report --method-sort=total --traces output.mlpd  > report.txt

And I generate the load on my test application using apache's ab


Also, I'm wondering if it's suppose to call
`System.Text.RegularExpressions.Interpreter:Eval
(System.Text.RegularExpressions.Interpreter/Mode,int&,int)` when
working with compiled regexs. My gut tells me it's not support to do
that, but I don't know the mono internals at all.

>
>
> On Mon, Dec 3, 2012 at 1:20 PM, Jordan Earls <earlz at lastyearswishes.com>
> wrote:
>>
>> So, I recently did a major update to my ASP.Net website running on
>> Mono (runs on .Net as well, but I work on it and host it with Mono).
>> Long story short, my websites performance sunk big time. It now takes
>> 110ms for a response that once took less than 20ms.
>>
>> I traced down the big difference to using regular expressions as a
>> part of my routing and a few other things. (I have a custom MVC
>> framework: https://bitbucket.org/earlz/barelymvc).
>>
>> Here are some(maybe all) of the regular expressions being used
>> (they're really simple):
>>
>> @"[^a-zA-Z0-9]\ "
>> @"[^a-zA-Z0-9\-]"
>> "^[0-9a-zA-Z]+$"
>> @"^[-+]?[0-9]*\.?[0-9]+$"
>>  "^[-+]?[0-9]*$"
>> "^[0-9A-Fa-f]+$"
>>
>> All of these regular expressions are compiled once at application
>> startup and cached. (with RegexOptions.Compiled). They are used with
>> with IsMatch(input) or Replace(input, replacement);
>>
>> When profiling my application however, the number one most time
>> consuming method was `System.Text.RegularExpressions.Interpreter:Eval
>> (System.Text.RegularExpressions.Interpreter/Mode,int&,int)` (my
>> database functions, which are also not really fast didn't even compare
>> to the amount of time this function consumed)
>>
>> Is there anything I can do to mitigate this problem? I'm using the
>> latest stable version of Mono (Arch Linux) on a 64-bit platform. Is it
>> true also that regular expression should now be on par with .Net's
>> implementation? From what I understand, mono compiles them to IL
>> correct? Is there anything to watch for in regular expressions that
>> could give me performance problems like this?
>>
>> Thanks,
>> Jordan
>> _______________________________________________
>> Mono-devel-list mailing list
>> Mono-devel-list at lists.ximian.com
>> http://lists.ximian.com/mailman/listinfo/mono-devel-list
>
>