Today we have two more findings.
1. The decrease in speed seems to come only if we create and drop clustered indexes, and then bcp-in to a table on the same device. If we bcp-in to a table on another device, speed is not suffering so mutch.
2. A collection of sysmonwatis.
SELECT msw.WaitTime,mwei.Description FROM monSysWaits msw inner join monWaitEventInfo mwei on msw.waitEventId = mwei.WaitEventId order by msw.WaitTime des
After starting some threads with bcp-in I ran this command, and after a few minutes I ran it again. This is the differences of the two.
| Description | WaitTime |
|---|---|
| xact coord: pause during idle loop | 720 |
| hk: pause for some time | 276 |
| waiting for incoming network data | 1028 |
| until an engine has been offlined wait | 180 |
| distributed xact co-ordinator idle loop | 180 |
| checkpoint process idle loop | 162 |
| Wait until heartbeat or check interval expires | 12 |
| waiting for client connection request | 236 |
| waiting on run queue after sleep | 417 |
| waiting on run queue after yield | 188 |
| waiting for last i/o on MASS to complete | 59 |
| waiting for network send to complete | 41 |
| wait for i/o to finish after writing last log page | 20 |
| wait to acquire latch | 19 |
| waiting for buf write to complete before writing | 12 |
| waiting for a lock | 15 |