Database Reference
In-Depth Information
Chapter 15
Additional Estimation
Considerations
ž Detailed look at the assumptions behind the QUBE
ž Nonleaf pages and leaf pages with regard to buffer pools, subpools, and
disk read cache
ž Occasions when the response time can be much less than the QUBE
ž Leaf and table pages remaining in memory
ž Small tables, hotspots, and comebacks
ž Skip-sequential
ž Calculating the CPU time
ž Default CPU coefficients
ž Measurements and considerations
ž CPU estimates used in making index design comparisons
ASSUMPTIONS BEHIND THE QUBE FORMULA
The simple QUBE formula is based on the following main assumptions:
1. CPU queuing time is insignificant—several processors, reasonable pro-
cessor load.
2. Disk drive utilization (drive busy) is between 15 and 35%.
3. Other queuing is insignificant.
4. CPU time is the dominant component of sort time.
5. Index nonleaf pages are in memory or in the disk server read cache, but
index leaf pages and table pages are always read from disk.
6. Processor and disk drive speeds:
ž At least 200 mips or 1 GHz per processor.
ž Disk drives 10,000 or 15,000 rpm, average seek time less than 5 ms.
ž
Sequential read from disk 40 MB/s.
Search WWH ::




Custom Search