Efficient, Distributed Data Placement Strategies for Storage Area Networks

ABSTRACT In the last ouple of years a dramati growth of enterprise data storage apa ity an be observed. As a result, new strategies have been sought that allow servers and storage being entralized to better manage the explosion of data and the overall ost of ownership. Nowadays, a ommon approa h is to ombine storage devi es into a dedi ated network that is onne ted to LANs and/or servers. Su h networks are usually alled storage area networks (SAN). A very important aspe t for these networks is s alability. If a SAN undergoes hanges (for instan e, due to insertions or removals of disks), it may be ne essary to repla e data in order to allow an eÆ ient use of the system. To keep the in uen e of data repla ements on the performan e of the SAN small, this should be done as eÆ iently as possible. In this paper, we investigate the problem of evenly distributing and eÆ iently lo ating data in dynami ally hanging SANs. We onsider two s enarios: (1) all disks have the same apa ity, and (2) the apa ities of the disks are allowed to be arbitrary. For both s enarios, we present pla ement strategies apable of lo ating blo ks eÆ iently and that are able to qui kly adjust the data pla ement to insertions or removals of disks or data blo ks. Furthermore, we study how the performan e of our pla ement strategies hanges if we allow to waste a ertain amount of apa ity of the disks. Resear h is partially supported by the DFG-Sonderfors hungsberei h 376 \Massive Parallelit at: Algorithmen, Entwurfsmethoden, Anwendungen." and the DFG/HNIGraduiertenkolleg \Parallele Re hnernetzwerke in der Produktionste hnik" yDepartment of Ele tri al Engineering and Information Te hnology, and Heinz Nixdorf Institute (HNI). Paderborn University. D-33102 Paderborn, Germany. zGraduiertenkolleg. HNI. Paderborn University. D-33102 Paderborn, Germany. xDepartment of Mathemati s & Computer S ien e, and HNI. Paderborn University. D-33102 Paderborn, Germany.