Name:Bittorrent over Grid
Description:Data center applications are still a challenging issue for large scale distributed
computing systems. The rise of new protocols and software for collaborative content
distribution over the Internet, offers a new opportunity for efficient and fast delivery for a
high volume of data. This work, supported by CO.ME.TA. infrastracture, presents an
evaluation of the bittorrent protocol. In particular, is proposed a new incentive mechanism,
in order to improve the overall system efficency in terms of bandwith, scalability and
fairness. To accomplish this, was developed a bittorrent simulator, using NS2, that is the
de-facto standard for network simulation. Thousands of experiments, using the
CO.ME.TA. grid infrastructure, were conducted in order to evaluate the potential of
bittorrent classical protocol compared to this new approach. The goal was to measure
basic charateristics of bittorrent in terms of latency, bandwidth and scalability for the
delivery of resources. Starting from these results, we plan to conduct new experiments
using the MPI technology available on the CO.ME.T.A. grid infrastructure. The design of
NS2 is such that simulation of very large networks is difficult and expensive, due to
excessive memory and CPU time requirements. We intend to use the MPI technology to
allow NS2 to be run in a parallel and distributed fashion, using the remote distributed
memory acess (RDMA) paradigm as well as the distributed cpu resources of the grid.
Message passing primitives and syncronization routines are the core idea, in order to allow
separate simulator instances communicating each others. By distributing the network
model on several machines, the memory requirements on any single system can be
substantially smaller than the memory used in a single-station simulation. The overall
execution time of the simulation should be at least as fast as the original single-station
simulation, since communications between istances relies onto the RDMA ion. We
also plan to investigate the porting of peer-
Abstract:Data center applications are still a challenging issue for large scale distributed
computing systems. The rise of new protocols and software for collaborative content
distribution over the Internet, offers a new opportunity for efficient and fast delivery for a
high volume of data. This work, supported by CO.ME.TA. infrastracture, presents an
evaluation of the bittorrent protocol. In particular, is proposed a new incentive mechanism,
in order to improve the overall system efficency in terms of bandwith, scalability and
fairness. To accomplish this, was developed a bittorrent simulator, using NS2, that is the
de-facto standard for network simulation. Thousands of experiments, using the
CO.ME.TA. grid infrastructure, were conducted in order to evaluate the potential of
bittorrent classical protocol compared to this new approach. The goal was to measure
basic charateristics of bittorrent in terms of latency, bandwidth and scalability for the
delivery of resources. Starting from these results, we plan to conduct new experiments
using the MPI technology available on the CO.ME.T.A. grid infrastructure. The design of
NS2 is such that simulation of very large networks is difficult and expensive, due to
excessive memory and CPU time requirements. We intend to use the MPI technology to
allow NS2 to be run in a parallel and distributed fashion, using the remote distributed
memory acess (RDMA) paradigm as well as the distributed cpu resources of the grid.
Message passing primitives and syncronization routines are the core idea, in order to allow
separate simulator instances communicating each others. By distributing the network
model on several machines, the memory requirements on any single system can be
substantially smaller than the memory used in a single-station simulation. The overall
execution time of the simulation should be at least as fast as the original single-station
simulation, since communications between istances relies onto the RDMA ion. We
also plan to investigate the porting of peer-
Created:2010-05-01
Last updated:2010-05-01