solaris performance question

kevin12 · Post by **kevin12** » Wed Oct 10, 2001 1:04 pm

We are trying to tune the performance of a texis instance runnning on a 420r with 4gb of RAM, RAID5, the main index is just over 2gb. The only other interesting process running on the box gets files from the network so they can be added to the table and indexed.

The copying process, which does a fair amount of disk I/O, seems to run up the search times considerably. Below are average search times for each hour of the day yesterday. At 11:00 I ran some extra file copying, which drove up the search times. From 15:00 to 20:00 the file copying was turned off, and search times got much better.

Our production box doesn't seem to have the same problem. The only differences that I can find are that it is RAID1+0 instead of RAID5, it has two functioning SCSI controllers instead of 1, and the copying process only maintains one file tree instead of two.

Do you guys have any experience with this, and can you suggest a way for us to approach it?

date hour avg search
011009:00 4.85
011009:01 6.05
011009:02 5.11
011009:03 7
011009:04 5.1
011009:05 6.95
011009:06 5.71
011009:07 5.58
011009:08 5.7
011009:09 6.21
011009:10 5.68
011009:11 11.47
011009:12 9.24
011009:13 7
011009:14 6.44
011009:15 4.22
011009:16 2.4
011009:17 1.79
011009:18 1.93
011009:19 1.77
011009:20 6.55
011009:21 3.95
011009:22 4.95
011009:23 4.12

Post by **John** » Wed Oct 10, 2001 2:06 pm

Is the RAID being done in software? Writing to RAID 5 does tend to be slower, and consume more resources as it needs to calculate the parity for every write. Raid 1+0 is a better solution. Also with 2 SCSI controllers you may have effectively twice the bandwidth, which certainly helps.

Was the search load the same on each box? The other impact that large copies can have is removing the database blocks from the disk cache. The higher the query rate to copy rate the more will stay cached, so if the query rate was higher on the production box it may have kept more data cached.