I'm not sure this route would work in a clustered environment-- or server restarts.
>I am not sure about this. You can always stick it in app scope. I am curious >about the work needed to keep the mapping data in sync with the component >config data, as there is a lot of work being done already in syncronized >regions of ApplicationImpl . > >Dennis Byrne > >>-----Original Message----- >>From: Martin Marinschek [mailto:[EMAIL PROTECTED] >>Sent: Wednesday, June 7, 2006 06:52 AM >>To: 'MyFaces Development' >>Subject: Re: change state saving algorithm for tree structure >> >>Sounds good to me - but where do you store long - component class name >>mapping? >> >>regards, >> >>Martin >> >>On 6/6/06, Dennis Byrne <[EMAIL PROTECTED]> wrote: >>> >>> The tree structure makes up about one third of the state for state. As >>> those of you who were involved in gathering test data for our presentation >>> at JavaOne already know, state serialization is a performance killer for >>> client side state saving. >>> >>> Currently MyFaces and the RI store component.getClass().getName() and >>> component.getId() for the tree structure. Does anyone think it would be >>> better to map the class name to a simple short ? You could perform the >>> mapping at startup as each component is configured. That's a lot less to >>> serialize and encrypt. It's better on bandwidth also. Not too difficult >to >>> implement. Have I missed anything here? >>> >>> Dennis Byrne >>> >>> @Martin, I implemented the parallelized state saving, very small but very >>> consistent improvement. >>> >>> >>> >> >> >>-- >> >>http://www.irian.at >> >>Your JSF powerhouse - >>JSF Consulting, Development and >>Courses in English and German >> >>Professional Support for Apache MyFaces >> > >