On 10/12/2007, Xiao-Feng Li <[EMAIL PROTECTED]> wrote: > > On Dec 10, 2007 5:02 PM, Simon Chow <[EMAIL PROTECTED]> wrote: > > Yes, I agree that. > > > > But I think MVM's feature of sharing class runtime representations > across > > applications' can make many apps benefit not only at startup phases, but > > also at runtime. > > Yes. Runtime sharing could be useful. Without MVM, this can be made > available through OS supports to dynamic shared objects (with careful > design).
good idea! > Besides these, I guess processes scheduling may be useful when running 2 > > related programs like server and database. > > And migration Java processes between defferent JVMs would be useful for > load > > balance and reliability in distributed enviroment. > > I remember MVM was only for shared memory system, no support to > distributed environment (unless you meant distributed shared memory > system, but that's too far a concept). I means that maybe we can migrate a running "Java process" in a heavily loaded JVM to a free JVM in another computer. This may be similar to real process migration in different OS, AFAK, which is difficult to implement. But for Java, I think it will be more easily than that. This is only my thought, maybe far from MVM. But it could be based on MVM. Btw, please use the word > "process" carefully, because the tasks in MVM are not the processes in > traditional sense. My understanding of "process" is clearly the entity > with address space isolation. Thank you, I will remember that. :=) > Ok, it may be useless idea in practice. But if all these things can be > > achived just by MVM. Is it valuable to do it? > > The problem is, whether we want to achieve those benefits with MVM if > they can attained through other approaches. In my opinion, the key > concept of MVM is to share the same virtual address space between > multiple JVM instances. But since the address space isolation with > process has been well established, proven, and mature in current > platforms, then how to share the space needs deep thinking and serious > proving. I saw the TODO Comments in method JNI_CreateJavaVM like this: //TODO: only one VM instance can be created in the process address space. Dose this mean what you said above? :) I am interested how to do these just when I saw the comments. Simon, it's good for you to think of these things. It's always > encouraged. And my comments are only my personal opinions, just for > your reference. > > Thanks, > xiaofeng Thank you Xiaofeng! Btw, I heard about you from HuangBo when I told him I am working on Harmony after his adv. compiler class. What a small world! > Thanks for your comments! > > > > > > On 10/12/2007, Xiao-Feng Li < [EMAIL PROTECTED]> wrote: > > > > > > On Dec 10, 2007 2:49 PM, Simon Chow <[EMAIL PROTECTED]> wrote: > > > > Acturally, I am just a greenhand for JVM. > > > > After seeing some paper about MVM, which says MVM can reduce the > startup > > > > > > > overhead and footprint of programs, I think both destop application > and > > > > server application can benefit from it. > > > > In my opinion, at least, some IDE can use it for debugging small > program > > > > more conveniently by avoiding startup overhead. > > > > > > > > Could you say something about its shortage? > > > > Thank you! > > > > > > Simon, I think the advantage (shorter startup time) of MVM you > > > mentioned can be achieved without MVM. Don't you think so? > > > > > > Thanks, > > > xiaofeng > > > > > > > > On 10/12/2007, Xiao-Feng Li <[EMAIL PROTECTED]> wrote: > > > > > > > > > > On Dec 10, 2007 1:00 PM, Simon Chow < [EMAIL PROTECTED]> > wrote: > > > > > > Will harmony support the multi-tasking feature? > > > > > > I has seen that Sun has a project name > > > > > > Barcelona< http://research.sun.com/projects/barcelona>which > aimed to > > > > > > run several Java applications on 1 JVM. > > > > > > Maybe it can improve the JVM performance somehow. > > > > > > I am very interested about this, but I can not find more > resource > > > about > > > > > > this.... > > > > > > > > > > In foreseeable future, Harmony might have no plan for MVM support. > I > > > > > had some investigations in this area, and I personally am not > fully > > > > > convinced by the idea. > > > > > > > > > > Thanks, > > > > > xiaofeng > > > > > > > > > > > -- > > > > > > From : [EMAIL PROTECTED] School of Fudan University > > > > > > > > > > > > > > > > > > > > > > > > > > -- > > > > > http://xiao-feng.blogspot.com > > > > > > > > > > > > > > > > > > > > > -- > > > > > > > > From : [EMAIL PROTECTED] School of Fudan University > > > > > > > > > > > > > > > > -- > > > http://xiao-feng.blogspot.com > > > > > > > > > > > -- > > > > From : [EMAIL PROTECTED] School of Fudan University > > > > > > -- > http://xiao-feng.blogspot.com > -- >From : [EMAIL PROTECTED] School of Fudan University
