Bogdan Nicolae, Bogdan.Nicolae@irisa.fr
GNU Lesser General Public License (LGPL) version 3.
This software is available on INRIA's forge. Registration with AP P is in progress.
BlobSeer is a huge blob (binary large object) management service to be used as a specialized storage backend for large scale distributed computing applications that abstract data input as huge sequences of bytes (such as MapReduce applications). It exports a simple, yet versatile blob manipulation interface, that allows reading, writing and appending to huge blobs while providing full transparency with regard to data allocation, replication and fault tolerance. BlobSeer has been developed since January 2008 within the JuxMem INRIA Gforge project, and has moved in its separate project since September 2008. The current implementation features 23,000 lines of C++/Perl. The service itself relies on the Boost collection of C++ libraries, libconfig and OpenSSL. Perl code is used to handle deployment on Grid'5000 , which is done through the OAR resource scheduler. Preliminary tests have proven correctness and performance with up to 400 nodes from 3 different sites. Ongoing development targets integration with Hadoop and PostgreSQL