[Catalyst] Non-real world irrelevant benchmarks

[Michael Reece]

fwiw, earlier this week i ran some simple 'ab -n 100 -c 10'  
benchmarks comparing an existing complex Mason site with the same  
site sitting behind a simple Catalyst front end.

the results showed nearly identical performance, with Mason-only  
winning slightly for 100 requests, and Catalyst-powered winning  
slightly for 200 requests.

and by slightly, i do mean slightly -- 3.37 req/s vs 3.40 req/s in  
one test, 3.30 req/s vs 3.26 req/s in the other test.  (slowish,  
sure, but that's just a sandbox server and the /index.html page  
tested is very heavily database driven.)

at this point, the Catalyst front end simply replicates the  
functionality previously done via custom mod_perl handlers (setting  
up cobrands, databases, etc), letting the default controller pass  
through to the (custom) Mason view.

these benchmarks were enough to easy management worries about the  
recent Rails/Django/Catalyst benchmarks, and we are proceeding with  
refactoring our application into Catalyst with the expectation that  
we will eventually see better performance by having many functions  
currently handled by Mason be handled by Catalyst controllers instead  
(such as dynamic image creation, form processing and redirect-after- 
POST, authentication/authorization management).

michael.

On Nov 16, 2006, at 6:01 PM, Perrin Harkins wrote:

> Nilson Santos Figueiredo Junior wrote:
>> Anyone knows how Mason compares to TT, performance-wise?
>
> There are some very old benchmarks here:
> http://chamas.com/bench/#2000
>
> These are not ideal though, because they compare the cost of using  
> Mason as your controller and templating system to the cost of a  
> simple mod_perl handler plus TT.  What you really want is a  
> benchmark comparing them through Catalyst.
>
> Regarding your TT performance issues, there are some common things  
> to do, which you can find in the TT list archives.  I'd still like  
> to see your dprof output sorted by real time, since it seems  
> impossible that a TT template could take 4 seconds unless it's  
> doing something like fetching from a database or the TT cache is  
> turned off.
>
> - Perrin
>