Re: [GNUnet-developers] forwarding of p2p queries

[Christian Grothoff]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Tuesday 29 April 2003 08:45 am, Glenn McGrath wrote:
> Im trying to track down what i see as an excessive amount of p2p queries
> initiated from my node.
>
> To track it down i setup 3 fresh nodes isolated from the rest of the
> net, let them idle for a while and made sure they knew about each other,
> then restarted them to reset the stats.
>
> node 1: no content, initiate search, and let it go for 40 minutes.
> node 2: has content that is searched for.
> node 3: no content, observer node.
>
> I do a search request from node1, look for content i know is on node2,
> these are the results.
>
>
> Test A uses both udp and tcp transport.
>
> node 1:
> # p2p queries sent     : 7
> # p2p queries received : 14
>
> node 2:
> # p2p queries sent     : 7
> # p2p queries received : 15
>
> node 3:
> # p2p queries sent     : 7
> # p2p queries received : 14
>
>
> Test B uses only tcp transport, and only went for 1/2 as long.
>
> node 1:
> # p2p queries sent     : 6
> # p2p queries received : 10
>
> node 2:
> # p2p queries sent     :  6
> # p2p queries received : 11
>
> node 3:
> # p2p queries sent     :  5
> # p2p queries received : 12
>
>
> Firstly, these numbers dont simply add up.

Check where the counteres are incremented. One thing may be that the p2p 
queries sent counter is incremented by one regardless of how many peers will 
be targeted (that would explain (most of) the difference: one query send to 
two recipients). Also, if you are not exclusively looking at search queries, 
super-queries may be only counted once at the sender but up to 25 times at 
the receiver -- again, depending on where the stats-calls are.

> Test A:
> Total of 21 p2p queries sent, 43 recieved.
>
> Test B:
> Total of 17 p2p queries sent, 33 recieved.
>
> In both tests there were about twice as many recieved queries as sent
> queries, where did the extra queries come from ?

Again, I would suspect the position of *where* the counters are is the main 
issue. Plus-minus one query can easily happen since you'll have a hard time 
getting the stats at all peers at exactly the same moment.

> I initialy suspected that the queries were being sent on tcp and udp
> (hence the 2 tests), but that is shown to be not the case.
>
> The only other thing i can think of is that 1 p2p query sent is like a
> mutlicast that is recived by both the other nodes, doesnt sonnd right
> though.

Well, it's not a multicast on the wire. But that the code may receive one 
query and forward it to two other peers (and maybe only count it as one 
query?) should be possible. Maybe we'll want to move the place where whe 
increment the counter...

> The other unexplained behaviour is that the observer node behaved the
> same as the node doing the query in the first test, but not in the
> second. But maybe that was just a timing error.
>
> Is the routing method described in one of the papers. i had a quick look
> and didnt see it.

It's described a little bit in the GAP paper, but not in too much detail. I 
have not gotten around to putting it into writing.

Christian
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.6 (GNU/Linux)
Comment: For info see http://www.gnupg.org

iD8DBQE+rr5e9tNtMeXQLkIRAvODAJ9CPa4NLw/DCFUeC+4w6p2UlDtvPgCeIc3/
dFrUIoaC3a5AtcNzplBG4S4=
=a32+
-----END PGP SIGNATURE-----