|
Windows 2000 Out-of-the-Box Sequential Buffered SCSI Throughput
|
| bet | 21/47 | | Sana | 30.03.2021 | | Hajmi | 1,33 Mb. | | #13789 |
|
Next we measure the performance of Windows 2000 when sequentially reading and writing SCSI disks. Riedel’s graph of NT4SP3, the left graph of Figure 13, is of measurements taken on a slower machine with slower disks. Compared with Riedel’s machine, the “new” test machine’s processors have more than seven times the number of available clock cycles between the two of them and the “new” drives are almost 2.5 times faster. We can learn much however from comparing Riedel’s graph to the current measurements since although the vertical scale has changed, the trend lines remain the same. NT4SP3 shows a much greater disparity between WCE writes and reads.
Basic and dynamic volumes are shown in the middle and right graphs of Figure 13. They have almost identical throughputs. Both buffered reads and writes show good performance for all request sizes. Particularly good performance was achieved at small request sizes as FS buffering coalesced the requests into 64KB blocks. This allowed the small requests to run at disk speed. The only exception to this rule was the case of 2KB writes where the throughput was slightly less than 11.5MBps with WCE and slightly less than 7MBps without WCE. This is since, for each 2KB write, Win2K reads a 4KB page off disk and then merges the 2KB write with it. The 4KB read before each 2KB write more than doubles the overhead for each request. Even with this additional overhead, buffered 2KB writes still offered a throughput advantage over unbuffered 2KB writes. Increasing the depth of buffered requests resulted in minimal performance gains.
|
| |
