On 30 Jul 2006, [EMAIL PROTECTED] wrote:
> I've got my GnutellaNet settings to limit connections of 25% to
> anyone vendor (and reserve 50% for GTKG nodes). However I always
> seem to end up with all Limewire connections (in bold).
>
> Is this because satellite nodes (the greyed out ones) count towards
> the totals? I don't seem to be seeing any other GTKG nodes.

I don't think that GTKG nodes are that common.  Crawler statistics or
something else might be useful so that people could gauge whether this
is an anomaly or not.  When I run as an Utlra with 25-40 nodes, I can
get about 3-4 GTKG ultra peers after about three days.

However, I do notice a distinct preference to either Limewire or
Bearshare ultras.  It seems to depend on build/host cache, with either
vendor becoming dominant.  It might be that they are only recommending
hosts from the same vendor?

I do believe that leaf nodes (satellites) count towards GTKG nodes.  I
wouldn't set the limits as you have.  I think that 10-20% for GTKG is
sufficient.  Also, you might want to increase your vendor limit to at
least 33%.  I think that 40% might be better as most nodes seem to be
LimeWire and BearShare.  It takes some time to collect ultras not from
these vendors.

It was also recently discussed to decrease the connection limit (seen
at the bottom of the GNET pane) to a lower value like 10-20.  This
limits the amount of "guesses/probes" that GTKG makes to find these
hosts.  These probes take a lot of CPU.

It would make some more sense to use some sort of "PID" algorithm to
try and target the vendor limits.  For example, the minimum ultra is
set to 40 and the vendor limit is 50%.  If during connect a vendor has
20 connection, all similar vendors are dropped even if there are 20
slots left to fill.  It might make more sense to keep half of them (20
left to be filled).  As the original 30 ultra nodes are lost they will
be filled with a different vendor.  So after a long time, GTKG would
reach the vendor target but more ultras would be available sooner
possibly helping to populate the host cache with more current
information.  [Actually the algorithm above is just "proportional",
with no "integral" or "derivative" portion].

Another "win" in my opinion would be to store the vendor id in the
host cache.  It doesn't seem to make sense to probe hosts that
previously had a vendor id that is over the current limit.  Providing
a field for a DNS entry would also be helpful.

Of course all this would take extra coding.  And perhaps I have a
limited understanding of the difficulties and sublties since I have a
very immature understanding of GTKG.

fwiw,
Bill Pringlemeir.


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys -- and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Gtk-gnutella-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/gtk-gnutella-devel

Reply via email to