Bayou Experiment 2   
=============================================================
28Feb07

In this experiment, we try to show that if a small device only subscribes
to what it is interested in, it greatly improves the anti-entropy
performance

Node 1 starts of with a populated DB
Node 2 invokes entropy for 
   - full db
   - for 10% of the data


There are 2 different sizes for which the experiments were run
a 100K db -- the results are in Exp2_combinedResults.txt

 Obj size   number	Total	%	path (each dir has 10% of the data)
 10b	      3000	 30K	30%	/10/1, /10/2, 10/3
 100b	      400	 40K	40%	/100/1,/100/2, /100/3, /100/4
 1K	      30	 30K	30%	/1000/1, /1000/2, /1000/3
 
and  a 1M db -- the results in Exp2_big_combinedResults.txt

  Obj size   number	Total	%	path (each dir has 10% of the data)
  10b 	   10,000	100k	10%	/10/1
  100b	   2,000	200K	20%	/100/1, /100/2
  1K	   400		400K	40%	/1000/1,/1000/2, /1000/3, /1000/4
  10K	   20		200K	20%	/10000/1, /1000/2
  100K	   1		100K	10%	/100000/1
 

In both cases Node 2 subscribe for /100/1/* directory

The results are plotted as averages of 8 iterations

Note:
- For this experiment we used a hack which does not send bound invals of objs not in the SS and sends imprecise invals for those (InvalIterator.java)
- Also we set the the FORCE paramter in OutgoingConnection.java to 10 so that it will send the PI as soon as it is ready.



