Hi,

The Enterprise site starting adding new files to download.  I have not
yet been able to get one of them.  After five or six tries I can usually get the
process started but none have managed to get 50% of the blocks in the first
segment tried.  With 0.51 all the files were downloadable...

My node is getting between 800 and 1200 requests per hour, with out bound 
bandwidth limited to 6000.  It has not needed to announce since May 12th (after 
it was down 24 hours for a computer shuffle).

Here are some items from my nodes stats:

Histogram of requested keys.
This count has nothing to do with keys in your datastore
May 21, 2003 10:02:24 PM
keys: 1711
scale factor: 0.16161616146564484 (This is used to keep lines < 64 characters)

   0 |===============================================================
   1 |========================================
   2 |===============
   3 |========
   4 |===
   5 |===
   6 |==========
   7 |================
   8 |===============================
   9 |============================
   a |======
   b |==================
   c |=====
   d |=======
   e |=======
   f |=======


Histogram of keys in in fred's data store
These are the keys to the data in your node's local cache (DataStore)
May 21, 2003 10:02:47 PM
keys: 21913
scale factor: 0.025692492723464966 (This is used to keep lines < 64 characters)

   0 |==================================
   1 |==================================
   2 |=========================================
   3 |===============================
   4 |===========================
   5 |=========================
   6 |============================
   7 |=========================
   8 |=====================================
   9 |==============================================
   a |===================================
   b |==============================================
   c |=======================
   d |==========================
   e |===============================================================
   f |================================


Probability of success of an incoming request
May 21, 2003 10:03:19 PM
0 | 0.032581452
1 | 0.06666667
2 | 0.030303031
3 | 0.02
4 | -
5 | -
6 | 0.045454547
7 | 0.03883495
8 | 0.030612245
9 | 0.039106146
a | 0.09756097
b | 0.033898305
c | -
d | 0.044444446
e | 0.04347826
f | 0.0

Notice that the most requested keys are not near my datastore's specialization 
and the 
most successful keys do not match either of the above histograms.  The
datastore has not been reset since January, its oldest key is from Feb 25th.  
This
should be long enough for freenet to figure out what is in it - this seems not 
to be
happening.

I track the unstable freenet build and am currently running 6041 with Sun java 
1.4.1_02
on linux.

Conversations on irc have mentioned that others are also seeing problem .   I 
suspect 
what might be happening is that nodes drop out of the routing table before my 
node
is able to collect enough references on them to make smart routing choices.  
For instance, 
under the 'key count' column I see only 10 of 50 nodes with 40 or more refs, 
about 25 
have under 10 refs (9 with only 1 ref) - my node knows basicly _nothing_ about 
half the nodes 
in its routing table...

One thing that might help this is a much large routing table.  We could limit 
the nodes
we route to the to top N nodes.  A node would move into the top N (maybe N 
could be 
dynamic) when we see a reference to it as a datasource.  This way we could 
conserve the 
refs for nodes but not incur the overhead of routing to too many nodes.  
Another item to 
think about is that nodes seem to drop off my routing table when the 
'Successful 
Connections' reads between 40-60%.  This means that a loaded node could easily 
be lost 
(with its valuable refs).  Interesting one person for whom freenet was working 
for the above 
files was using 100 nodes with up to 64 refs per node.

Ed Tomlinson
_______________________________________________
devl mailing list
devl at freenetproject.org
http://hawk.freenetproject.org:8080/cgi-bin/mailman/listinfo/devl

Reply via email to