Declustered disk array architectures with optimal and near-optimal parallelism

Abstract

This paper investigates the placement of data and parity on redundant disk arrays. Declustered organizations have been traditionally used to achieve fast reconstruction of a failed disk's contents. In previous work, Holland and Gibson identified six desirable properties for ideal layouts; however; no declustered layout satisfying all properties has been published in the literature.We present a complete, constructive characterization of the collection of ideal declustered layouts possessing all six properties. Given that ideal layouts exist only for a limited set of conjigurations, we also present two novel layout families. PRIME and RELPR can tolerate multiple failures in a wide variety of configurations with slight deviations from the ideal. Our simulation studies show that the new layouts provide excellent parallel access perjormance and reduced incremental loads during degraded operation, when compared with previously published layouts. For large accesses and under high loads, response times for the new layouts are typically smaller than those of previously published declustered layouts by a factor of 2.5.

Permutation development data layout (PDDL)

“…In a declustered disk array [1,2,6,7,10], the number of stripe units per stripe is smaller than the total number of disks. Any usable data layout (the assignment of stripe units to stripes) fulfills a number of fairly natural criteria:…”

“…Recent declustering schemes PRIME of Alvarez, Burkhard, Stockmeyer and Cristian [2] and DATUM of Alvarez, Burkhard, and Cristian [1] also use straightforward, efficient calculation to map stripe units to disks thereby avoiding table lookup. Previous schemes include Parity Declustering of Holland and Gibson [6] which uses table lookup to specify balanced incomplete block design together with parity rotation.…”

Optimizing large data transfers in parity-declustered data layouts

“…This tradeoff of additional redundant space for reduced failure recovery time can be achieved using a technique called parity declustering, in which the stripe size k is chosen to be smaller than the array size v. Parity-declustered data layouts have been considered by, among others, Muntz and Lui [12], Holland and Gibson [10], Stockmeyer [20], Schwabe and Sutherland [16,18], Schwabe, Sutherland, and Holmer [19], Alvarez, Burkhard, and Cristian [1], and Alvarez, Burkhard, Stockmeyer, and Cristian [2].…”

“…Thus the effective use of the available parallelism for large transfers of data was not being guaranteed. Alvarez, Burkhard, Stockmeyer, and Cristian [2] formalized this idea by showing that it is impossible to satisfy both Conditions 4 and 5 simultaneously for most array configurations, where an array configuration is given by a pair (v, k) specifying the number of disks in the array and the stripe size. In fact, they showed that it is only possible when either k = v, k = v − 1, k = 2, or (v, k) is (5, 3), (7,3), or (7,4).…”

Proposta de um ambiente de simulação e aprendizado inteligente para RAID.

“…Estudos adicionais (Chen et al, 1990;Chervenak, 1990;Lee et al, 1991a;Lee et al, 1991b;Weikum, 1992;Stodolsky et al, 1993;Chen et al, 1994;Lee et al, 1996;Alvarez et al, 1998) mostram ou propõem características de desempenho e confiabilidade para situações específicas.…”