Linux Sun's JVM needs improvements

[Gain Paolo Mureddu]

    As I already pointed out in another thread, I really think the Linux 
JVM provided by Sun (even the latest 1.5.0) needs serious 
improvements... Whether it be applications such as Azureus (Bit Torrent 
client), LimeWire (p2p client), or web applets embeded in webpages, 
whenever Java loads in Linux, it is such a system hog!! I've seen it use 
as much memory as 400 Mb (according to top and ps) for an application 
that in itself should not use more than a couple tens of megs, like a 
p2p client... the problem is not the program which runs on top of Java, 
that usually is under expected memory ranges, but the Java process 
itself is the one that uses the most memory, plus it will begin to leak 
sometimes... Looks as if the Linux JVM had zero garbage collection.


Sun should really take care of thier JVM if they want to make Java 
succeed in *nix. Even their won JVM has many troubles running on their 
Solaris platform... So this is more a *nix in general issue than just 
Linux alone.

[Sam Varshavchik]

Gain Paolo Mureddu writes:

>  Sun should really take care of thier JVM if they want to make Java 
>  succeed in *nix. Even their won JVM has many troubles running on their 
>  Solaris platform... So this is more a *nix in general issue than just 
>  Linux alone. 
I agree 100%.

Unfortunately, it looks like you posted your message to the wrong mailing 
list.  This is not the Sun Java mailing list.

[Gain Paolo Mureddu]

Sam Varshavchik wrote:

>  Gain Paolo Mureddu writes:
>
> > Sun should really take care of thier JVM if they want to make Java 
> > succeed in *nix. Even their won JVM has many troubles running on 
> > their Solaris platform... So this is more a *nix in general issue 
> > than just Linux alone.
>
>
>  I agree 100%.
>
>  Unfortunately, it looks like you posted your message to the wrong 
>  mailing list.  This is not the Sun Java mailing list.
>  
This was actually a clean follow up to a series of posts about memory 
management in FC and Java came up, hence I forked the thread (too long, 
anyway) into this thread. So no, it is not the wrong list ;)

[Edward Yang]

Gain Paolo Mureddu wrote:

>  Sam Varshavchik wrote:
>
> > Gain Paolo Mureddu writes:
> >
> >> Sun should really take care of thier JVM if they want to make Java 
> >> succeed in *nix. Even their won JVM has many troubles running on 
> >> their Solaris platform... So this is more a *nix in general issue 
> >> than just Linux alone.
> >
> >
> >
> > I agree 100%.
> >
> > Unfortunately, it looks like you posted your message to the wrong 
> > mailing list.  This is not the Sun Java mailing list.
> >
>  This was actually a clean follow up to a series of posts about memory 
>  management in FC and Java came up, hence I forked the thread (too 
>  long, anyway) into this thread. So no, it is not the wrong list ;)
>  
Okay, I want to add more about Sun's JVM performance.

I have posted a message to Sun's forum about the memory footprint of the 
JVM on FC3. Please visit 
http://forum.java.sun.com/thread.jspa?threadID=590956 for the message.

[Gain Paolo Mureddu]

Edward Yang wrote:

>  Okay, I want to add more about Sun's JVM performance.
>
>  I have posted a message to Sun's forum about the memory footprint of 
>  the JVM on FC3. Please visit 
>  http://forum.java.sun.com/thread.jspa?threadID=590956 for the message.
>
>  
I was about to point out that you were talking about *Linux*, in this 
case Fedora Core 3 distribution, and everyone (apparently) undestood 
that you were using *Windows* and that the problem showed in Windows... 
I don't know if you could try and run a dedicated session of FC, just to 
make sure the results are consistent with what you get in VPC... 
Personally I've run into >400Mb memory foot print of Java *only* with 
Azureus (a famous Java Torrent client), other applications, such as 
LimeWire while DO populate the RAM get about 3/4 of that... However 
*any* Java applet run through Firefox or Konqueror will *at the very 
least* load 80Mb into memory, this after a fresh boot with a base memory 
usage of about 98 Mb (measured with GkrellM)... I'll try to post in 
general Linux forums about this in other distributions specificially 
with Sun's JVM.

[James McKenzie]

Gain Paolo Mureddu wrote:
>  Edward Yang wrote:
>  
> > Okay, I want to add more about Sun's JVM performance.
> >
> > I have posted a message to Sun's forum about the memory footprint of 
> > the JVM on FC3. Please visit 
> > http://forum.java.sun.com/thread.jspa?threadID=590956 for the message.
> >
> >
>  I was about to point out that you were talking about *Linux*, in this 
>  case Fedora Core 3 distribution, and everyone (apparently) undestood 
>  that you were using *Windows* and that the problem showed in Windows... 
>  I don't know if you could try and run a dedicated session of FC, just to 
>  make sure the results are consistent with what you get in VPC... 
>  Personally I've run into >400Mb memory foot print of Java *only* with 
>  Azureus (a famous Java Torrent client), other applications, such as 
>  LimeWire while DO populate the RAM get about 3/4 of that... However 
>  *any* Java applet run through Firefox or Konqueror will *at the very 
>  least* load 80Mb into memory, this after a fresh boot with a base memory 
>  usage of about 98 Mb (measured with GkrellM)... I'll try to post in 
>  general Linux forums about this in other distributions specificially 
>  with Sun's JVM.
>   
Let us know the results.  I may go to Sun and see if there is a thread 
about this.

-- 
James McKenzie

[Jim Cornette]

Edward Yang wrote:
>  Gain Paolo Mureddu wrote:
>  
> > Sam Varshavchik wrote:
> >
> >> Gain Paolo Mureddu writes:
> >>
> >>> Sun should really take care of thier JVM if they want to make Java 
> >>> succeed in *nix. Even their won JVM has many troubles running on 
> >>> their Solaris platform... So this is more a *nix in general issue 
> >>> than just Linux alone.
> >>
> >>
> >>
> >> I agree 100%.
> >>
> >> Unfortunately, it looks like you posted your message to the wrong 
> >> mailing list.  This is not the Sun Java mailing list.
> >>
> > This was actually a clean follow up to a series of posts about memory 
> > management in FC and Java came up, hence I forked the thread (too 
> > long, anyway) into this thread. So no, it is not the wrong list ;)
> >
>  Okay, I want to add more about Sun's JVM performance.
>  
>  I have posted a message to Sun's forum about the memory footprint of the 
>  JVM on FC3. Please visit 
>  http://forum.java.sun.com/thread.jspa?threadID=590956 for the message.
>  
>   
So if I understand this a bit, you are running Windows and then using 
Microsoft virtual machine to run Fedora, then running a java virtual 
machine and it is being a resource hog?

Did you try to run RHL 8.0, then launch the microsoft virtual machine 
program through wine to launch Fedora and see if the memory resources 
are any better?

Jim


-- 
Freedom is nothing else but the chance to do better.
		-- Camus

[Edward Yang]

Jim Cornette wrote:

>  Edward Yang wrote:
>
> > Gain Paolo Mureddu wrote:
> >
> >> Sam Varshavchik wrote:
> >>
> >>> Gain Paolo Mureddu writes:
> >>>
> >>>> Sun should really take care of thier JVM if they want to make Java 
> >>>> succeed in *nix. Even their won JVM has many troubles running on 
> >>>> their Solaris platform... So this is more a *nix in general issue 
> >>>> than just Linux alone.
> >>>
> >>>
> >>>
> >>>
> >>> I agree 100%.
> >>>
> >>> Unfortunately, it looks like you posted your message to the wrong 
> >>> mailing list.  This is not the Sun Java mailing list.
> >>>
> >> This was actually a clean follow up to a series of posts about 
> >> memory management in FC and Java came up, hence I forked the thread 
> >> (too long, anyway) into this thread. So no, it is not the wrong list ;)
> >>
> > Okay, I want to add more about Sun's JVM performance.
> >
> > I have posted a message to Sun's forum about the memory footprint of 
> > the JVM on FC3. Please visit 
> > http://forum.java.sun.com/thread.jspa?threadID=590956 for the message.
> >
> >
>
>  So if I understand this a bit, you are running Windows and then using 
>  Microsoft virtual machine to run Fedora, then running a java virtual 
>  machine and it is being a resource hog?
>
>  Did you try to run RHL 8.0, then launch the microsoft virtual machine 
>  program through wine to launch Fedora and see if the memory resources 
>  are any better?
>
>  Jim
>
>  
Sorry, I think there is some misunderstanding about virutal techonology.

In a virutal hardware box, the guest OS inside it does not know where it 
is. It may be a litter or much slower than in a real hardware box, but 
the memory footprint should never been any difference. That's why many 
developers use Microsoft Virutal PC or VMWare to test there programs. 
Think about it - you don't have to boot into another OS or go to another 
computer that may be 50 meters far just to test a small problem of your 
progarm!

So I think I I have FC3 installed in a real hardware box, the memory 
footprint of the JVM will be the same.

[Robin Laing]

Edward Yang wrote:

>  Sorry, I think there is some misunderstanding about virutal
> techonology.
>  
>  In a virutal hardware box, the guest OS inside it does not know where it 
>  is. It may be a litter or much slower than in a real hardware box, but 
>  the memory footprint should never been any difference. That's why many 
>  developers use Microsoft Virutal PC or VMWare to test there programs. 
>  Think about it - you don't have to boot into another OS or go to another 
>  computer that may be 50 meters far just to test a small problem of your 
>  progarm!
>  
>  So I think I I have FC3 installed in a real hardware box, the memory 
>  footprint of the JVM will be the same.
>  
>   
And we can all trust Microsoft to ensure that this is how it works. 
Microsoft lost a very long court case after making Windows not work 
with Dr Dos.  I wouldn't be surprised to find Microsoft doing 
something similar.  Who really knows all the code in Windows.


-- 
Robin Laing
Instrumentation Technologist   Voice: 1.403.544.4762
Military Engineering Section   FAX:   [phone]
Defence R&D Canada - Suffield  Email: Robin.Laing(a)DRDC-RDDC.gc.ca
PO Box 4000, Station Main      WWW:http://www.suffield.drdc-rddc.gc.ca
Medicine Hat, AB, T1A 8K6
Canada

[Jeff Vian]

On Tue, 2005-02-01 at 11:44 -0700, Robin Laing wrote:
>  Edward Yang wrote:
>  
>  > Sorry, I think there is some misunderstanding about virutal techonology.
>  > 
>  > In a virutal hardware box, the guest OS inside it does not know where it 
>  > is. It may be a litter or much slower than in a real hardware box, but 
>  > the memory footprint should never been any difference. That's why many 
>  > developers use Microsoft Virutal PC or VMWare to test there programs. 
>  > Think about it - you don't have to boot into another OS or go to another 
>  > computer that may be 50 meters far just to test a small problem of your 
>  > progarm!
>  > 
>  > So I think I I have FC3 installed in a real hardware box, the memory 
>  > footprint of the JVM will be the same.
>  > 
>  > 
>  
>  And we can all trust Microsoft to ensure that this is how it works. 
>  Microsoft lost a very long court case after making Windows not work 
>  with Dr Dos.  I wouldn't be surprised to find Microsoft doing 
>  something similar.  Who really knows all the code in Windows.
>  
>   
Especially since Linux is the thorn in Microsoft's side right now.

[Edward Yang]

Robin Laing wrote:

>  Edward Yang wrote:
>
> > Sorry, I think there is some misunderstanding about virutal techonology.
> >
> > In a virutal hardware box, the guest OS inside it does not know where 
> > it is. It may be a litter or much slower than in a real hardware box, 
> > but the memory footprint should never been any difference. That's why 
> > many developers use Microsoft Virutal PC or VMWare to test there 
> > programs. Think about it - you don't have to boot into another OS or 
> > go to another computer that may be 50 meters far just to test a small 
> > problem of your progarm!
> >
> > So I think I I have FC3 installed in a real hardware box, the memory 
> > footprint of the JVM will be the same.
> >
> >
>
>  And we can all trust Microsoft to ensure that this is how it works. 
>  Microsoft lost a very long court case after making Windows not work 
>  with Dr Dos.  I wouldn't be surprised to find Microsoft doing 
>  something similar.  Who really knows all the code in Windows.
>
>  
Well, I think your reply does not have much relation to virtual 
techonology even m$ is very evil.

[Robin Laing]

Edward Yang wrote:
>  Robin Laing wrote:
>  
> > Edward Yang wrote:
> >
> >>
> >
> > And we can all trust Microsoft to ensure that this is how it works. 
> > Microsoft lost a very long court case after making Windows not work 
> > with Dr Dos.  I wouldn't be surprised to find Microsoft doing 
> > something similar.  Who really knows all the code in Windows.
> >
> >
>  Well, I think your reply does not have much relation to virtual 
>  techonology even m$ is very evil.
>   
It does if you thing that MS could change one bit of code and 
distribute it in an update that would cause some weird problems for 
Linux in a VM environment.

What did they change in Windows to prevent it from installing on Dr. Dos?

How about all the fuss saying that they could not remove media player 
from Windows as it was integrated into the OS.  Now they are about to 
release a version of Windows without Media Player in Europe.  Or cut 
down versions of Windows for third world countries?

As Linux is becoming a bigger thorn in Microsofts side, they will try 
what they can to stop the spread or usage.  They are putting pressure 
on manufacturers to sell all computers with an OS installed and even 
charging for each computer sold even if it doesn't include Windows. 
All under the guise of stopping piracy.

VM's are a great idea but they are not clean installs and the 
underlying OS can still cause problems.  If the underlying OS is 
closed and controlled by a threatened company, they could do something 
to cause problems and blame it on a bug due to a security update.  And 
we all know that many bugs in Windows take along time to get fixed.

How many years did the battle go on with Dr. Dos.  Could the industry 
wait that long for Linux to work properly in a VM again?

In my opinion, with the cost of computers taking a dive, it is cheaper 
to have a second computer for these tests and use a KVM.  That is what 
we do around my work.  Heck, some people have three or four computers 
under/on their desks.

-- 
Robin Laing

[Jim Cornette]

Edward Yang wrote:
>  Jim Cornette wrote:
>  
> > Edward Yang wrote:
> >
> >> Gain Paolo Mureddu wrote:
> >>
> >>> Sam Varshavchik wrote:
> >>>
> >>>> Gain Paolo Mureddu writes:
> >>>>
> >>>>> Sun should really take care of thier JVM if they want to make Java 
> >>>>> succeed in *nix. Even their won JVM has many troubles running on 
> >>>>> their Solaris platform... So this is more a *nix in general issue 
> >>>>> than just Linux alone.
> >>>>
> >>>>
> >>>>
> >>>>
> >>>>
> >>>> I agree 100%.
> >>>>
> >>>> Unfortunately, it looks like you posted your message to the wrong 
> >>>> mailing list.  This is not the Sun Java mailing list.
> >>>>
> >>> This was actually a clean follow up to a series of posts about 
> >>> memory management in FC and Java came up, hence I forked the thread 
> >>> (too long, anyway) into this thread. So no, it is not the wrong list ;)
> >>>
> >> Okay, I want to add more about Sun's JVM performance.
> >>
> >> I have posted a message to Sun's forum about the memory footprint of 
> >> the JVM on FC3. Please visit 
> >> http://forum.java.sun.com/thread.jspa?threadID=590956 for the message.
> >>
> >>
> >
> > So if I understand this a bit, you are running Windows and then using 
> > Microsoft virtual machine to run Fedora, then running a java virtual 
> > machine and it is being a resource hog?
> >
> > Did you try to run RHL 8.0, then launch the microsoft virtual machine 
> > program through wine to launch Fedora and see if the memory resources 
> > are any better?
> >
> > Jim
> >
> >
>  Sorry, I think there is some misunderstanding about virutal techonology.
>  
>  In a virutal hardware box, the guest OS inside it does not know where it 
>  is. It may be a litter or much slower than in a real hardware box, but 
>  the memory footprint should never been any difference. That's why many 
>  developers use Microsoft Virutal PC or VMWare to test there programs. 
>  Think about it - you don't have to boot into another OS or go to another 
>  computer that may be 50 meters far just to test a small problem of your 
>  progarm!
>  
>  So I think I I have FC3 installed in a real hardware box, the memory 
>  footprint of the JVM will be the same.
>  
>   
You might try these tests using a real set of hardware. I did tests 
using Vmware to convert a legacy machine up to the latest. The test 
indicated that everything was fine.
I don't think that the virtual hardware is truely imitating real 
hardware in all aspects.
I would trust tests using real hardware more than tests on virtual 
machines. I have not had much trouble installing fedora on real 
machines, but too much using virtual machines.

Jim

[Rob Saul]

On Jan 31, 2005, at 3:41 PM, Gain Paolo Mureddu wrote:

>     As I already pointed out in another thread, I really think the 
>  Linux JVM provided by Sun (even the latest 1.5.0) needs serious 
>  improvements... Whether it be applications such as Azureus (Bit 
>  Torrent client), LimeWire (p2p client), or web applets embeded in 
>  webpages, whenever Java loads in Linux, it is such a system hog!! I've 
>  seen it use as much memory as 400 Mb (according to top and ps) for an 
>  application that in itself should not use more than a couple tens of 
>  megs, like a p2p client... the problem is not the program which runs 
>  on top of Java, that usually is under expected memory ranges, but the 
>  Java process itself is the one that uses the most memory, plus it will 
>  begin to leak sometimes... Looks as if the Linux JVM had zero garbage 
>  collection. 
I use Eclipse everyday as a development tool and to my knowledge it's 
never approached 400Mb
is size.  Given it's appetite for resources I'd tend to suspect that an 
app taking so much more memory is poorly
written.  The VM can't make up for bad programming.

Just the same,  Sun's port to Linux came as an after thought and rode 
on the shoulders of the Blackdown
team ( www.blackdown.org ).  I think Sun gives more lip service to 
Linux than actual engineer time.

>  Sun should really take care of thier JVM if they want to make Java 
>  succeed in *nix. Even their won JVM has many troubles running on their 
>  Solaris platform... So this is more a *nix in general issue than just 
>  Linux alone. 
you might try posting your thoughts to the java-linux list ( see 
blackdown mentioned above ).
I know at least one Sun engineer reads it regularly and responds to 
issues.