- This topic has 11 replies, 1 voice, and was last updated 8 years, 6 months ago by Kalman.
May 21, 2015 at 8:02 pm #918Anas AnisGuest
According to the theoretical description of GIA, the replication rate
has some effects on the performance of the overlay, but how to control
this rate in the simulator?May 21, 2015 at 8:02 pm #919Matthias FeldottoGuest
At the moment only an one-hop replication of documents to all neighbors is implemented. Therefore all nodes store replicated content from neighbors locally. Instead of one-hop there are many strategies (for example with a parameter for the replication rate) imaginable, but not yet implemented. Feel free to implement them yourself if you want to use it.
As entry point you find the implemented strategy in the class de.tud.kom.p2psim.impl.overlay.gnutella.gia.OneHopReplicator.May 21, 2015 at 8:03 pm #920Anas AnisGuest
I thought the replication rate is something different from the one-hop
replication! I thought the replication rate is the replication of
documents inside the overlay, but the one-hop replication is like
pointers of the direct neighbors’ contents and not the contents
themselves. so the replication rate has nothing to do with neighbors!
(this is what I understood)May 21, 2015 at 8:03 pm #921Anas AnisGuest
So am I right?May 21, 2015 at 8:04 pm #922Anas AnisGuest
another question concerning GIA, i’m getting results of ‘average hop
counts’ when simulating this overlay around 0.5. What is the meaning
of having 0.5 as number of hops? I expected the number of hops to be
at least 1 or more.May 21, 2015 at 8:04 pm #923Matthias FeldottoGuest
The implemented One-Hop replication provides (java class comment)
* Component that manages the one-hop replication of documents to all neighbors and stores replicated
* content from neighbors locally.
I’m not deep in the topic of GIA, but another replication is not yet implemented. So if you need other replication you have to implement it yourselfMay 21, 2015 at 8:10 pm #924Matthias FeldottoGuest
The average hop count could also be less than 1, because if the key is managed at the local node there will be a hop count of 0 for this request.May 21, 2015 at 8:10 pm #925Anas AnisGuest
Thanks for your reply.
I have another question, but this time about Gnutella0.4. How can I
simulate the failure (or leaving) of Gnutella0.4 nodes inside the
I’m using GnutellaApplication as the componentClass in the
configuration file. In GnutellaApplication there is no fail or leave
functionality to be called. So how to simulate such a scenario?May 21, 2015 at 8:11 pm #926Matthias FeldottoGuest
To simulate the leaving of hosts you have two possibilities:
1. Let a node leave the network by using the leave() or fail() methods
of the GnutellaOverlayNode class (These operations are implemented not
in the application because they are direct functions of p2p nodes).
2. You can use a ChurnGenerator (see different example configurations).
It provides the leaving and joining of nodes depending on different
Depending on your scenario one of these options should help you.May 21, 2015 at 8:11 pm #927Anas AnisGuest
I would like to use the first possibility because the second (churn)
doesn’t offer me a permanent leave of nodes (I think it works like
restarting nodes). But how to use these functions if I’m using
GnutellaApplication? or can I use the GnutellaOverlayNode instead of
GnutellaApplication? I searched for an example like that, but I didn’t
find, and I tried to make my own example and I failed because I
couldn’t write an action for connecting nodes.
ThanksMay 21, 2015 at 8:12 pm #928Matthias FeldottoGuest
You can use methods from more than only one class in your action file.
Look for an example in the configuration
“config/application-filesharing2/FSChord-32_Peers.xml” of the current
simulator’s version. The classes are named by the attributes
“componentClass” and “additionalClasses”. You can use methods in the
actions file with “class:method” (compare the actions-file for the
example).May 21, 2015 at 8:12 pm #929KalmanGuest
Churn offers also a the option for permanent leave, in the case the
the rate of joining nodes is lower than the rate of failing nodes.
As Matthias said, Churn is affecting the network layer of the nodes,
not the application or overlay layer, so you should check the commands
for the network layer. Have also a look at the Chord example, as it
demonstrates how to set up the commands to make nodes leave your
network. However, it is necessary to digg into the code to see how the
actions in the actions file take effect in order to understand, how to
use them properly.
- You must be logged in to reply to this topic.