Author Topic: Problems with Sun's Java Virtual Machine  (Read 2643 times)

0 Members and 1 Guest are viewing this topic.

Offline Nemo

  • Global Moderator
  • Elite
  • *****
  • Posts: 1303
  • Karma: +27/-0
    • View Profile
Problems with Sun's Java Virtual Machine
« on: July 03, 2004, 03:57:24 PM »
Source: http://article.gmane.org/gmane.network.freenet.devel/14363
[code:1]
From: Toad <toad@...>
 Subject: I hate Sun
 Newsgroups: gmane.network.freenet.devel
 Date: Wed, 30 Jun 2004 21:03:27 +0100
 
<iip> <nickster> i am using Sun's 1.4.2 JRE, because nothing else works
as reliably
<toad__> wonderful
<toad__> 1.4.2 has MAJOR bugs
<toad__> it will not work well
<toad__> having said that, you can say the same of 1.4.1 probably
<toad__> I use 1.4.1
<iip> <nickster> well, they claim to have fixed major 1.4.1 bugs in
1.4.2
<toad__> yes
<toad__> however they have introduced some major bugs too
<toad__> in the BigInteger support
<toad__> these affect routing...
<toad__> we don't get misrouting, we just get NullPointerExceptions and
NumberFormatExceptions in JVM code...
<toad__> which generally can't be worked around
<iip> <nickster> i use 2.6 kernel and NPTL is very nice
<iip> <nickster> hmm, can't freenet have its own BigInteger module
<toad__> I use 2.6 too
<toad__> freenet has far too much of its own stuff duplicated because of
bugs in one or other JVM
<toad__> and it would be slower than the JVM impl
<iip> <nickster> with 1.4.1 jvm i had a separate process for each
thread, which was eating up mem for lunch
<toad__> the fix is to fix GCJ, or to use 1.4.1
<iip> <nickster> with 1.4.2 jvm it's better, but still not perfect
<iip> <nickster> (just one process now)
--
Matthew J Toseland - toad@...
Freenet Project Official Codemonkey - http://freenetproject.org/
ICTHUS - Nothing is impossible. Our Boss says so.

_______________________________________________
Devl mailing list
Devl@...
http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl
[/code:1]

The message thread on http://news.gmane.org/gmane.network.freenet.devel continues:
[code:1]From: Toad <toad@...>
 Subject: Re: I hate Sun
 Newsgroups: gmane.network.freenet.devel
 Date: Wed, 30 Jun 2004 21:41:35 +0100
 
<toad__> i don't see why it being just one "process" has ANYTHING to do
with RAM usage
<toad__> of course they share memory
<toad__> they just show up as multiple PIDs
<iip> <nickster> yeah, but it was sharing very little, actually
<toad__> no
<toad__> it shares 100.0000%
<toad__> there is a tiny bit of per thread
<toad__> but it's only a tiny bit
<toad__> you were misreading top
<toad__> it's very easy to misread top, top is very misreadable :|
<iip> <nickster> when my swap space was used up almost entirely (512mb),
that was not misreading
<toad__> I can't believe they'd have more than a tiny amount of per
thread...
<toad__> well that's not the reason
<toad__> it will be something else
<iip> <nickster> and my load was >20 from swapping
<iip> <nickster> i had nothing else running
<toad__> well, maybe 1.4.1 is a memory hog
<iip> <nickster> well mayve
<toad__> but large amounts of per process memory sounds very unlikely
<toad__> I mean it's certainly not going to duplicate actual node
structures
<iip> <nickster> well, perhaps jvm needed big stacks for each process
<toad__> well, it will need just as big a stack for each NPTL thread
<toad__> nickster: well, if you tell me 1.4.1 is a memory hog, then i
believe you
<toad__> but I get really bad problems with 1.4.1

--
Matthew J Toseland - toad@...
Freenet Project Official Codemonkey - http://freenetproject.org/
ICTHUS - Nothing is impossible. Our Boss says so.

_______________________________________________
Devl mailing list
Devl@...
http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl
[/code:1]

Perhaps the newer JVM solves the problems?
[code:1]
From: Someone <someone@...>
 Subject: Re: I hate Sun
 Newsgroups: gmane.network.freenet.devel
 Date: Thu, 01 Jul 2004 16:41:52 +0200
 
Toad schrieb:

> <toad__> however they have introduced some major bugs too
> <toad__> in the BigInteger support
> <toad__> these affect routing...
> <toad__> we don't get misrouting, we just get NullPointerExceptions and
> NumberFormatExceptions in JVM code...
> <toad__> which generally can't be worked around

With the 1.5 beta 2 I didn't see anything like this running unstable for about
12 hours. So, is it possible that those bugs are fixed there? Or was it just
luck?
[/code:1]
Does anyone see such problems in his JVM-Version?

Greetings,
Nemo.