[EMAIL PROTECTED] (Ian Clarke) writes:

> On Wed, Jun 05, 2002 at 04:32:20PM -0400, Andrew Rodland wrote:
> > arguments. Is anyone willing to try this? I'm still to java-incompetent
> > to code it, but if anyone wants to write it, I'd love to look at it. >:)
> 
> You should try to code it, if I am on #freenet I am happy to help.  
> Look at this as an opportunity to learn a-little Java - it isn't hard.
> 
The more people that understand the internals of the code, the better.

> My attitude is that load-balancing can hardly get worse, I am happy to 
> attempt anything that nobody can find a serious flaw with.
> 
> Ian.

I will have to spoil this attempt to add code to fred by pointing out
that nodes aren't removed from the routing table for having really low CP.
Nodes can have as bad a CP as they want and they'll only be bumped out
of if the table is full of nodes and a new reference comes in.

The problem is simply a single constant in CPAlgoData.java:

    private final static int MAXFAILUREINTERVALS = 6;

This means that when a node's failureIntervals (the number of
consecutive connection failures) is over 6, it's removed from the
routing table.  Six failure intervals take about an hour to rack up,
so if a node is down for an hour, it's completely eliminated.  I've
bumped this constant up to 60 on my node, basically eliminating the
possibility of removing a node from the routing table for being
uncontactable (the CP code eventually eliminates nodes when new refs
come in, and the exponential backoff makes sure I don't try to contact
them often at all.)

There are two ways to solve this; one quick fix and one real solution
that addresses (at least more of) the real problem:

1) change this constant to 12 or 15.  This will drastically increase
   the amount of time a node has to be down to be removed.  (because
   of the exponential backoff)

2) change the deletion code to just move references offline so that
   when the node is back up, all the info about it isn't lost.

Thelema
-- 
E-mail: [EMAIL PROTECTED]                        Raabu and Piisu
GPG 1024D/36352AAB fpr:756D F615 B4F3 BFFC 02C7  84B7 D8D7 6ECE 3635 2AAB

_______________________________________________
devl mailing list
[EMAIL PROTECTED]
http://hawk.freenetproject.org/cgi-bin/mailman/listinfo/devl

Reply via email to