Re: JRE as part of Linux

2001-01-27 Thread Tony Dean

Sounds what we need is the old UNIX sticky bit put back into Linux. The
sticky bit was a file attribute that told the Kernel that the
application should stay in cached for some period of time after it was
initially loaded. Way back when, UNIX did this and things that got hit a
lot like editors (vi, sed, etc) would stay cached in memory. Someting
like this could keep the JVM cached after it was initially started. The
initial startup could be something as simple as putting it in your
.profile, .login, or .bashrc with the instruction to print the version
upon initial login.

I'll drop in on one of the Kernel lists this week and see if there is
any interest.

Regards
Tony Dean

noisebrain wrote:
> 
> > What is really needed is a pre-started jvm.  When you start up a java
> > process, the jvm will fork, and the child will su to you and proceed as
> > normal.  I don't know exactly what the jvm is doing when it is taking
> > all that time starting up so I don't know how useful this would be.
> 
> I like this solution, though I don't see the details.
> My guess is that part of the startup time is just that java
> has to uncompress the classes zip file, which is big.  The
> scheme above would avoid the uncompress.
> 
> --
> To UNSUBSCRIBE, email to [EMAIL PROTECTED]
> with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]


--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: Fwd: Re: JRE as part of Linux

2001-01-27 Thread Nicolás Lichtmaier

> The problem is that the connection to the X display happens when the AWT
> native library is first loaded. It's a single connection that is used
> through the life of the JVM. As currently structured, the AWT won't let
> you associate different objects with different X connections. It would
> take some work (possibly deep architectural rework) on the AWT to do
> what you're asking.

 It can be done. There would be necessary to reimplement the Toolkit part of
the awt to make it perform calls to a forked JVM, that will handle each X
display. I think that IBM has something done about this, I think they call
it remote AWT and they use that to display AWT from GUI-less machines like
the AS/400.


--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]




Re: JRE as part of Linux

2001-01-27 Thread Doug Collinge

Oracle has developed a virtualized JVM for their various web products.  
It puts all the shared, read-only stuff in one layer and all the private 
read-write stuff in another layer.  It allows each client session to 
have effectively their own private JVM with something like 20KB overhead 
per session.  This seems to me a very promising way of having multiple 
independent JVMs on the same machine.  Solves the nastly little gotchas 
involved in sharing a JVM by giving you your very own JVM.

Anders Lindbäck wrote:

>> noisebrain wrote:
>> 
 What is really needed is a pre-started jvm.  When you start up a java
 process, the jvm will fork, and the child will su to you and proceed as
 normal.  I don't know exactly what the jvm is doing when it is taking
 all that time starting up so I don't know how useful this would be.
>>> 
>>> I like this solution, though I don't see the details.
>>> My guess is that part of the startup time is just that java
>>> has to uncompress the classes zip file, which is big.  The
>>> scheme above would avoid the uncompress.
>> 
> 
> It is also all the time needed to create the objects in the initial
> environemnt from the classes in the zip file. 
> 
> Tony Dean skrev:
> 
>> Sounds what we need is the old UNIX sticky bit put back into Linux. The
>> sticky bit was a file attribute that told the Kernel that the
>> application should stay in cached for some period of time after it was
>> initially loaded. Way back when, UNIX did this and things that got hit a
>> lot like editors (vi, sed, etc) would stay cached in memory. Someting
>> like this could keep the JVM cached after it was initially started. The
>> initial startup could be something as simple as putting it in your
>> .profile, .login, or .bashrc with the instruction to print the version
>> upon initial login.
> 
> 
> I´t wont help much since a large part are the classes in the zip file.
> 
> The way to fix that is if classes is in shared memory.
> Think of a deamon which has a prestarted JVM with thr base classes
> in read-only shared memory so all application could share them.
> 
> The deamon could run as root and whenever a user wants to start a new JVM 
> instead of creating a new JVM the java script will send of a request
> to the deamon. The deamon forks off a JVM with correct parameters and 
> set up to run under that users environment. 
> 
> It´s a prestarted JVM but requires some rewrite of how one can create
> JVMs. 
> 
> This deamon will in some ways work likea a kernel when it creates processes. 
> There are some xecurity problems if anyone may load new classes into the JVM
> but that can be solved in the same way as the sandbox moden in a web browser.
> 
> This would speed up JVM start time and also lower the memory requiremsnt
> when running JVM in different processes.
> 
> Anders
> 
> 
> --
> To UNSUBSCRIBE, email to [EMAIL PROTECTED]
> with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]



--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]