De-layered grid storage server
ACM SIGBED Review - Special issue: The second workshop on high performance, fault adaptive, large scale embedded real-time systems (FALSE-II)
40Gbps de-layered silicon protocol engine for TCP record
Proceedings of the conference on Design, automation and test in Europe: Proceedings
| Hi-index | 0.00 |
This document describes a distributed transaction driver developed to support the reimplementation of Swift with added RAID (Redundant Arrays of Inexpensive Disks) functionality. Both a high-level overview and a low-level program description are provided. The Swift system was developed to investigate the use of disk striping to achieve high I/O performance. The transaction driver described here has been used to implement RAID-0, RAID-4, and RAID-5 Swift systems. Swift uses a network of workstations in a manner similar to a redundant disk array, i.e., an application on a client node requests I/O via library routines which evenly distribute I/O across multiple server nodes. Data blocks in RAID files are distributed over the servers. RAID-0 contains no redundancy/parity information, RAID-4 uses a dedicated parity node, and RAID-5 uses distributed parity. The original Swift system used a straight-forward RAID-0 scheme that did not readily scale to RAID-4 and RAID-5 implementations. The transaction driver described here was developed to cope with the distributed concurrent programming problems posed by these implementations. In principle, this transaction driver can be used for a wide variety of distributed concurrent programming problems.