Re: [freenet-support] Re: What things of Freenet eats the most memory

2004-01-19 Thread Toad
On Tue, Jan 20, 2004 at 03:51:32AM +0100, Someone wrote:
 After more experiments with the node it seems like it needs way
 more memory than I can give it. I stopped all other services running
 on the machine and give fred the whole 192 MB that were not needed
 by windows. With this it managed to run 15 hours before running out
 of memory. So it seems to me that it is not possible to run fred
 on my machine. The node is now down and will stay down in the next
 time.

Try reducing the maximumThreads.
 
 It is not only the memory usage, the node simply needs to much
 attention. Because everytime it stalls it starts to eat up all
 CPU cycles and actually blocks/slows down the other services
 running on the machine. Reducing the priority of the node makes

What sort of machine are you running it on? I don't think it's a vital
project goal for fred to run well on a 300MHz machine.

 it stall even faster, so this is equal to stopping it. In addition
 the Traffic due the constant rising ammount of querys and connects
 did to not only DOS my node, it also started to DOS my whole
 internet connection and it took several hours till it normalized
 after shutting down the node. So it is not possible to simple switch
 of the node if I need the full bandwith of my connection for other
 things.
 
 I also can't understand why a program thats moving data from one
 peer to another needs this *big* ammount of memory and cpu cycles.

It probably could be optimized significantly w.r.t. memory. With regard
to both - there are two things. One is java - java is pretty inefficient
with memory usage, and always has been, and always will be. However on
the whole this is not a huge problem, and we cannot simply rewrite Fred
in C++/C/Perl/Python/Logo/Brainfart/Assembly, because it is ~ 140,000
lines of code, as well as for political and portability reasons. Another
is that Freenet is doing hard stuff, that is not clearly known to work,
the basic principles are still being established, and so on.

 I'm running a well used IRC bouncer, a private Webserver, a MySQL
 DB and more on the same machine. All of them together only take
 about 60 MB of memory and 10% to 15% cpu cycles when they are at
 their configured maximum usage at the same time (neighter memory
 nor cpu usage got higher than this in the last year). I do expect
 fred to need more because of routing and crypting but not that
 much that it needs now.

We can probably get Fred to use 100MB of RAM in the default config.
Eventually. However in the time it takes to get there, 100MB of RAM will
be so cheap that there will be little point.
 
 I know that freenet is still under development and that it is a
 really hard task to get something like this to a point where it
 is running really well (especially when using a slow ressource
 hog like java as base). So I'm keeping an eye on the mailing list,
 maybe there will be a break through in the near future making it
 possible to run fred on a machine like mine without it freaking
 out and beeing a major pain in the ass for my machine (I really
 hope so).
-- 
Matthew J Toseland - [EMAIL PROTECTED]
Freenet Project Official Codemonkey - http://freenetproject.org/
ICTHUS - Nothing is impossible. Our Boss says so.


signature.asc
Description: Digital signature
___
Support mailing list
[EMAIL PROTECTED]
http://news.gmane.org/gmane.network.freenet.support

[freenet-support] Re: What things of Freenet eats the most memory

2004-01-19 Thread Someone
Edward J. Huff schrieb:
5061 seems to have a memory leak.  I gave it 512M and it still 
got up to 450M (I restarted it so it wouldn't run out).  
At least I'm not the only one having this problem. I could live with
the node needing 128 MB memory if it will still work without stalling
when reaching the limit. Would it be possible to implement a automatic
node restart as soon as it has only 1 MB of free memory left. This
would at least make shure the node doesn't stall totally.
Routing remains an unsolved problem...  Read freenet-dev
on gmane for more info.
Yes, but it was never this bad QPH wise with earlier builds. With 5050
and earlier the QPH stabilized somewhere at 4 QPH. This was still
far more than my node could handle, but at least it didn't influence
my internet connection to a point where it is having similiar effects
to a DOS attack, that is even going on after the node has been down for
more than 1 hour. Especially since 5061 it has really gone mad.
QPH was at 15 when I took the node down.

-- Ed Huff
Greets someone

___
Support mailing list
[EMAIL PROTECTED]
http://news.gmane.org/gmane.network.freenet.support