Optimising disc head movements in secondary key retrievals

Estimating disk head movement in batched searching

“…This implies that the queries are satisfied by moving the disk heads forward starting from the first cylinder of the file. In 1-4] it is also proved that the total expected distance traveled by the disk head is : (1) ( N -1)q/(q+ 1) i.e. q/(q+ 1) of the portion of the file will be searched for satisfying the batch.…”

“…For example, if we assume that m = 2, then two possible queries are "Retrieve all employees where DEGREE = M.SC," and "Retrieve all employees where SALARY > 30000". In [1] it is observed that the total disk head movement may be reduced if instead of satisfying each query on a FCFS basis the sytem performs the index searches for all the m queries. The fact that the pointer part for each index is usually ordered in terms of the cylinder number, track number etc.…”

“…The analysis in [1] proceeds as follows. Let qi be the number of records satisfying each of the m queries (1 < i _< m).…”

“…For the rest of this study we make exactly the same assumptions made in [1,4]. In particular, we assume (a) that the file occupies N consecutive cylinders, (b) that the physical addresses on the records which satisfy a batch are sorted on ascending cylinder number and that the cost of sorting is negligible.…”

“…In both these earlier works [1,4] a replacement model was implicitly assumed and estimates for the expected cost were derived. The term replacement (nonreplacement) model implies that the probability for locating a record in a specific cylinder in the file remains (does not remain) constant when the cylinder has already been accessed for locating another record in the batch.…”