|
Windows 2000 Unbuffered Sequential SCSI Throughput
|
| bet | 24/47 | | Sana | 30.03.2021 | | Hajmi | 1,33 Mb. | | #13789 |
|
Figure 15 shows the almost identical performance between basic and dynamic volumes with WCE on one-deep unbuffered sequential requests. As synchronous requests are one deep from the standpoint of the disk, one-deep asynchronous and synchronous requests have the same results unless otherwise noted. We will show later in the paper that dynamic volumes have a slightly higher processor overhead than basic volumes. Despite this increased processor overhead, we did not see any effect on unbuffered throughputs on our test machine. This was due to the imbalance between the processor and the disks. On our unloaded machine, the processors are easily able to handle the increased workload of dynamic volumes. Regardless of whether basic or dynamic volumes are used, large requests with WCE combine to provide optimal performance.
A sizable 30% improvement is seen for 2KB reads on two-deep and greater read requests over one-deep requests. At one-deep read requests, the throughput is 17.7MBps, while at two-deep and above the throughput is just slightly under disk speed at 23MBps. This is due to additional pipelining: the amount of time that the disk sits idle is decreased. The computer works to try to ensure that one request is always ready for the disk. There was no improvement for increased write depth. The plateau on writes is an artifact of the disk drive and is explained later.
|
| |
