Next: RAID1 physical stripe placement
Up: Experimental Results and Analysis
Previous: Experimental Results and Analysis
The effect of asynchronous writes on the performance of TSS device can be
seen by comparing the RAID1, RAID5 and cRAID5 performance with synchronous
and asynchronous writes (with max 8 async writes at a time). This results are
summarized in Table 1.
The results indicate a significant improvement in performance of TSS, due to
asynchronous writes in case of RAID1 and a moderate improvement in case of
RAID5 and cRAID5. This improvement is achieved due to the reduction of
waiting time through the use of asynchronous writes. In case of asynchronous writes,
we have a separate thread running, which does the cleanup for each of the
request. The time taken by this cleanup operation is not included in the
results, as they happen asynchronously and they do not form a part of
response time for the requests.
All of the following trace runs have been taken with async
writes enabled in TSS.
Table 1:
Effect of asynchronous writes on the performance of TSS (trace data from 6 [/var])
Type of |
Avg. write time |
Avg. I/O time |
Device |
per block (ms) |
per block (ms) |
|
sync |
async |
sync |
async |
|
writes |
writes |
writes |
writes |
RAID1 |
17.23 |
8.79 |
12.66 |
7.43 |
RAID5 |
21.98 |
19.32 |
14.87 |
13.47 |
cRAID5 |
94.87 |
74.36 |
65.90 |
54.00 |
|
Next: RAID1 physical stripe placement
Up: Experimental Results and Analysis
Previous: Experimental Results and Analysis
2001-09-13