hi people, 

We (Ian, Matthew and I) did some simple testing on the
ResponseTimeEstimator (RTE) object (see Matthew's
implementation at 
http://cvs.sourceforge.net/cgibin/viewcvs.cgi/freenet/freenet/src/freenet/node/rt/ResponseTimeEstimator.java?rev=1.3&content-type=text/vnd.viewcvs-markup
). 
We had one-day real-world traffic logged on the node
hawk.freenetproject.org:xxxx, 
and used only those successful request information in
the simulations.

Specifically, we picked from the logged traffic two
nodes showed frequently: nodeA:48828 (113 successful
requets), and nodeB:1313 (73 successful requests).
We then ran RTE on the traffic from those two nodes
individually, using the following methodology:
For each request, the RTE object was firstly asked to
estimate the response time for the given key. Then, it
was given the real response time to update its
database.
The response time is defined as (search time +
transfer time for transferring average-size file from
the next-hop node).

We called the above "full-training" testing and
expected the estimation error decrease along with the
time (the # of requests fed into the RTE object).

The results are ploted in graphs at 
 
 
http://netweb.usc.edu/~huizhang/New_topic/NG_Routing/keyspace_nodeA.ps
http://netweb.usc.edu/~huizhang/New_topic/NG_Routing/rte_nodeA.ps
 
 
http://netweb.usc.edu/~huizhang/New_topic/NG_Routing/keyspace_nodeB.ps
http://netweb.usc.edu/~huizhang/New_topic/NG_Routing/rte_nodeB.ps
 
Three observations:
 1. The response time curve produced by RTE is closely
the "average" line of real response time. At this
point,
 we could say RTE achieved its original design goal.
 2. Even for the same node, the response time varied
 widely with a factor of 100.
 3. As for the estimation error of RTE, it was not
obvious to see the decreasing trend along   with the
time. We thought it's due to the much variation in
routing times shown in the  collected data, which is
hoped to be alleviated with the deployment of new
NG-Routing scheme.  
 
 
Discussion is highly welcome! As a start, I have two
questions here:
1. What should be the right initial value of
estimation time in RTE objects? Now it's 0, but I
think the right way is assigning a large response time
(i.e., if this object had no experience in this area,
he should say no).

2. How to include the failed request information into
the RTE? An intutive way is treating a failed request
as a successful request with a large response time.


 

Thanks,
 
Hui


__________________________________
Do you Yahoo!?
Yahoo! Calendar - Free online calendar with sync to Outlook(TM).
http://calendar.yahoo.com
_______________________________________________
devl mailing list
[EMAIL PROTECTED]
http://hawk.freenetproject.org:8080/cgi-bin/mailman/listinfo/devl

Reply via email to