Abstract
Expecting a high data delivery rate as well as data protection, the Lister Hill National Center for Biomedical Communications procured a RAID system to house image files for image delivery applications. A study was undertaken to determine the configuration of the RAID system that would provide for the fastest retrieval of image files. Average retrieval times with single and with concurrent users were measured for several stripe widths and several numbers of disks for RAID levels 0, 0+1 and 5. These are compared to each other and to average retrieval times for non-RAID configurations of the same hardware. Although the study in ongoing, a few conclusions have emerged regarding the tradeoffs among the different configurations with respect to file retrieval speed and cost.
Rationale and goals
The Lister Hill National Center for Biomedical Communications, a research and development division of the National Library of Medicine, procured a Sun SPARCstorage Array (SSA), model 101, to house image files for prototype image delivery applications. The SSA model 101 is configured with eighteen Seagate ST31200W 1.05 GB disks connected to six internal fast wide SCSI busses. The SSA is connected to a Sun SPARCstation 20 via a Fiber Channel port. SPARCstorage Volume Manager software supports use of the SSA as independent volumes or as:
RAID 0: Data is split into equal sized blocks, or stripes, and distributed among the disks in the RAID volume.
RAID 1: All data in a volume are duplicated on the mirror volume.
RAID 0+1: Both the original volume and the mirror volume are striped.
RAID 5: In Addition to data blocks, RAID Level 5 includes parity blocks, which are distributed among the disks in the RAID volume
[1,2].
The specifications of the Seagate disks
[3] in the SSA cite a data transfer rate of 3.3 to 5.9 MB/sec. The fast wide SCSI interface has a data transfer rate of 20 MB/sec, and the Fiber Channel connector has a data transfer rate of 25 to 50 MB/sec. Those specifications the following statements from a technical white paper led us to expect very high data retrieval rates in addition to the data security available from RAID.
- "Each of the disks in a stripe are generally assumed to be on their own independent data channel, allowing the transfer rate of a RAID 0 implementation to approach the sum of the transfer rates of each of the drives." [4]
- "Both SPARCstorage Array models ... are capable of over 2000 two-KB input-output operations per second, and sustained transfer rates exceeding 15 MB/second."[4]
