Argonne National Laboratory

Managed GridFTP

TitleManaged GridFTP
Publication TypeConference Paper
Year of Publication2011
AuthorsBresnahan, J, Link, M, Kettimuthu, R, Foster, IT
Conference NameProc. 2011 IEEE International Symposium on Parallel and Distributed Processing Workshops and PhD Forum
Date Published06/2011
Other NumbersANL/MCS-P1847-0211

GridFTP extends the standard FTP protocol to provide a high-performance, secure, reliable data transfer protocol optimized for high-bandwidth wide-area networks. The Globus GridFTP implementation has become the preeminent high-performance data transfer tool for the Grid community. Its modular architecture enables a standard GridFTP-compliant client access to any storage system that can implement its data storage interface, including the HPSS archival storage system, SRB, the GPFS parallel file system, and POSIX file systems. Its extensible I/O interface allows GridFTP to target high-performance wide-area communication protocols such as UDT, FAST TCP, and RBUDP. The Globus GridFTP server implementation already implements the concept of "striping," where multiple data movers are aggregated as one logical resource. However, there exists no mechanism in GridFTP to manage the use of server resources by the clients. When many transfer sessions occur simultaneously, the GridFTP server can overwhelm the transfer host and/or the underlying storage system. Moreover, there is no flexibility in the management of data movers in striped configuration. The data movers must be statically configured, and they cannot come and go dynamically. In this paper, we present a framework to manage the GridFTP resources efficiently so as to avoid overburdening host resources, prevent client starvation, and enable dynamic addition or removal of data movers.