[EMAIL PROTECTED] wrote: > Philippe, > > I did some tests and after all I now know that there are only small things that >could be improved in JOnAS (from my point of > view): Most time (90%) is used for pushing the result of find* back to the client >via RMI, even if client and server are on > the same machine. This is one of the main advantages of Jeremie over RMI : local calls are performed locally and thus do not involve any serialization. What you lose using Jeremie is distributed garbage collection (not available with RMI/IIOP anyway). > [...] > Or do you think using RMI-IIOP will perform in a faster way? There is no reason why RMI/IIOP should be really faster than RMI. The need to interface some RMI/IIOP elements with Java serialization (for instance, to handle externalizable objects) has a certain performance cost. On the other hand, GIOP (the IIOP methiod invocation protocol), provides a way to implement some sort of streaming, allowing recipients to start decoding messages before the sender has finished emitting data. This may have a very positive effect for really distributed applications. About local optimization (avoiding coding/decoding when the client and server are collocated), the specification states that arguments and results should be copied anyway, therefore losing a great portion of the expected advantage. The RMI/IIOP implementation that will be used by future releases of Jonas may be parameterized to avoid any copy. Best regards, Bruno
begin:vcard n:Dumant;Bruno tel;cell:06 75 20 76 64 tel;fax:33 1 49 26 09 76 tel;work:33 1 42 44 40 74 x-mozilla-html:FALSE url:www.kelua.com org:Kelua SA adr:;;55 rue Sainte Anne;Paris;;75002;France version:2.1 email;internet:[EMAIL PROTECTED] x-mozilla-cpt:;1 fn:Bruno Dumant end:vcard
