| 7-018 | |
| Peer-to-Peer Computing for Secure High Performance Data Copying | |
| Andrew Hanushevsky, SLAC and Artem Turnov, SLAC | The BaBar Copy Program (bbcp) is an excellent representative of peer-to-peer (P2P) computing. It is also a pioneering application of its type in the P2P arena. Built upon the foundation of its predecessor, Secure Fast Copy (sfcp), bbcp incorporates significant improvements performance and usability. As with sfcp, bbcp uses ssh for authentication; providing an elegant and simple working model -- if you can ssh to a location, you can copy files to or from that location. To fully support this notion, bbcp transparently supports 3rd party copy operations. The program also incorporates several mechanism to deal with firewall security; the bane of P2P computing. To achieve high performance in a wide area network, bbcp allows a user to independently specify, the number of parallel network streams, tcp window size, and the file I/O blocking factor. Using these parameters, data is pipelined from source to target to provide a uniform traffic pattern that maximizes router efficiency. For improved recoverability, bbcp also keeps track of copy operations so that an operation can be restarted from the point of failure at a later time; minimizing the amount of network traffic in the event of a copy failure. Here, we preset the bbcp architecture, it¡¯s various features, and the reasons for their inclusion.. ¡¡ |
| Keywords: | Performance, Networking, Data Copy |
| Contact: | Mr. Andrew Hanushevsky |
| Stanford Linear Accelerator Center | |
| abh@stanford.edu | |