Disk Backup Through Algebraic Signatures in Scalable Distributed Data Structures

A Scalable Distributed Data Structure (SDDS) allows to store a large scalable file over a distributed RAM. The file scales up transparently for the application over the nodes of a multicomputer, e.g., a local network of PCs. The prototype system termed SDDS 2000, designed by CERIA experiments with this technology for Wintel multicomputers. The application may manipulate data much faster than on local disks. We present the functions we have added to SDDS-2000 to backup the RAM on each storage node onto the local disk. Our goal is to write only the data that has changed since the last backup. We experiment for this purpose with the algebraic signatures. We present our architecture and design choices. Performance measures validate our implementation. It is now available for the noncommercial use as part of new version of our prototype termed SDDS-2002. 1. Introduction A Scalable Distributed Data Structure (SDDS) allows to store a large scalable file over a distributed RAM. The file scales up transparently for the application over the storage nodes of a multicomputer, e.g., a local network of PCs and WSs. The storage nodes are called the (SDDS) server nodes. The application manipulates through the SDDS interface at its node, called the (SDDS) client node. The client node performs the (scalable) address calculus and handles the application requests accordingly. Various SDDS schemes have been investigated, [C]. Probably the most studied were the linear hash partitioning variants of the LH* scheme, and the range