Hi
just some figures from my profiling sessions for verification (meaning:
have sou seen similar/comparable numbers)
Application: VERY simple application (crossbar helloWorld) with no
business
logic, just one string attribute with a pure
getter/setter-access
Servlet-Engine: Tomcat 5.0
Hardware: IBM R50 laptop (1.5MHz Centrino, 1GB of memory)
Profiler: Yourkit 4.0
JDK: Sun 1.4.2
Usecase: results in 10 requests, 1 with a validation exception
Load: 1 single user
Versions:
- MyFaces 1.1.0
- RI 1_1_01
(all values in milliseconds)
Lifecycle-phase
State Rest Apply Proc. Upd. Invok Rendr Total CPU
View Req. Valid. Model Appl Resp
ms/Request
MF Srvr 50 10 10 10 0 320 42
Clnt 300 10 0 0 10 741 108
RI Srvr 40 90 50 0 10 570 77
Clnt 711 70 40 20 20 1492 237
retained memory after 10 requests
MyFaces Server: 64kB (4 kB in Session)
Client: 65kB (0.5 kB in Session)
RI Server: 46kB (9 kB in Session)
Client: 29kB (0.5 kB in Session)
Size of transferred state-info (2 views...)
welcome.jsf response.jsf
MyFaces (2Fields) 1836 Bytes 1532 Bytes
RI (1 Field) 1444 Bytes 1220 Bytes
Conclusions:
- MyFaces seems to be lighter on CPU (roughly 50% of RI)
- RI seems to be lighter on Memory (roughly 50-65% of MF)
- MyFaces-server-state produces a smaller session (roughly 50% of RI)
- RI-client-state transfers less (about 75% of MF)
- RI has less memory allocated outside the session (roughly 50-65% of
MF)
Unknowns:
- What happens with the memory-footprint when more than 1 concurrent
user
are producing load?
- How are the CPU-footprints when the views get more complex?
Comments/Ideas/Recommendations are VERY welcome
Alexander
PS: if someone creates a facelet-version of the sample-application I
would
try to recreate the same measurements. Please contact me. The
Crossbar-
Application had to be changed to work with JSF 1.1!
-----Original Message-----
From: Martin Marinschek [mailto:[EMAIL PROTECTED]
Sent: Friday, December 02, 2005 12:37 PM
To: MyFaces Discussion
Subject: Re: Re: MyFaces for production?
Hi,
no performance measurement here - but 5 JSF intranet and 3 internet
applications deployed here on MyFaces, no problems with performance
whatsoever.
Interesting for the performance is the switch between client-side and
server-side state saving - client-side is better for memory,
server-side is much better for CPU performance!
regards,
Martin
On 12/2/05, Jesse Alexander (KBSA 21)
<[EMAIL PROTECTED]> wrote:
> Sounds like time to do thread-dumps (and -analysis)...
>
> Have you two done some performance-measurements on your apps?
>
> regards
> Alexander
>
> -----Original Message-----
> From: news [mailto:[EMAIL PROTECTED] On Behalf Of Werner Punz
> Sent: Friday, December 02, 2005 9:59 AM
> To: [email protected]
> Subject: Re: MyFaces for production?
>
> Francesco Consumi wrote:
>
> > Performance is good, we have only some problem with
> > http://www.istitutodeglinnocenti.it, our main site, that somewhere
> locks
> > java process at 100% of cpu, but we're investigating....
> >
> Interesting, I never have had that issue, with now two big JSF
programs
> an a number
> of small internal hacks.
> My guess is that there is some deadlock in the app code, to my
knowledge
> MyFaces does not have any isse in this area.
>
>
>
--
http://www.irian.at
Your JSF powerhouse -
JSF Consulting, Development and
Courses in English and German
Professional Support for Apache MyFaces
