[GNUnet-developers] Re: [Help-gnunet] INDIRECTION_TABLE_SIZE

[Christian Grothoff]

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

On Friday 06 September 2002 12:33 pm, Christian Drechsler wrote:
> hi!
>
> hm, i increased the indirection table to 131072 and it still was full
> after some seconds of gnunetd running. is that because i am downloading a
> very large file myself? IIRC in gnunet every 1024-byte-block must be
> queried for, right? so very big files will soon fill up the indirection
> table - blocking out others ... ?

Unlikely, the indirection table has some cleverness with respect to slot 
allocation. 

> how is that implemented? is there some kind of a query queue so that new
> queries come to turn when there is room in the indirection table again? or
> are they just discarded? what about query priority when the table is full
> already?

If the slot in the table is used, the new query is *currently* just discarded. 
Priority and ttl are considered when gnunetd decides if the slot is full (it 
may decide to throw out the previous entry and replace it).

> can an aggressor fill up the indirection table of a node completely by
> sending queries he knows there will be no answer to to the node at a time
> when he knows there is no high netload at that node, so he will get in
> even with zero priority? (What a sentence! X-) )

Sure, he gets in with zero priority, but his slot gets kicked out as soon as 
anything with priority arrives. I don't claim that the heuristic for the 
routing/slot allocation code is anywhere near perfect (and it may certainly
be a bottleneck), but increasiong INDIRECTION_TABLE_SIZE like crazy 
is not the solution.

Some of this code is still on my list of things to revise for 0.4.9, so I'll 
probably report back on this once we've got the revised code :)

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

iD8DBQE9eO8B9tNtMeXQLkIRAgbvAJ9WV24HpDKqC03kLt2NJSISvdyYQACfeQ0T
9bfFmRgDME1I2Bz9ZYxBnO4=
=kRIi
-----END PGP SIGNATURE-----