1994 Fiscal Year Final Research Report Summary
Practical Studies on Cautious Schedulers for Object-Oriented Database Concurrency Control
| Project/Area Number |
05680282
|
|---|
| Research Category |
Grant-in-Aid for General Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
計算機科学
|
|---|
| Research Institution | KOBE UNIVERSITY OF COMMERCE |
|---|
Principal Investigator |
KINIWA Jun Kobe University of Commerce, Dept.Management Science, Associate Professor, 商経学部, 助教授 (90177882)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
KATOH Naoki Kobe University of Commerce, Dept.Management Science, Professor, 商経学部, 教授 (40145826)
|
|---|
| Project Period (FY) |
1993 – 1994
|
| Keywords | concurrency control / two-phase locking / cautious scheduler / long transaction / object-oriented database / deadlock / serializability / consistency |
|---|
| Research Abstract |
We evaluated the performance of every known cautious schedulers by simulation study, which was issued as a working paper No.137 by Institute of Economic Research, Kobe University of Commerce. These experiments, commonly based on a short transaction model, are classified by (a) single-version vs.multiversion, (b) the newest k-version vs.dynamically selected k-version, (c) recovery control vs.non-recovery control, and (d) other recent schedulers. As a result, we should choose the schedulers with dynamic constraints rather than the one with multiple versions. We can avoid inefficiency even if our recovery control method is used in the short transaction model. This contribution covers our first plan, that is, to investigate the performance of the cautious scheduler with recovery control.
Our next goal, in 1994, was how to control concurrently long transactions among short ones. If a long transaction, frequently used in object-oriented database systems, issues lock requests in two-phase fashion, its long locked period blocks many short transactions. We propose a cautious two-phase locking scheme in which short transactions have priority over long ones with probability p. We estimated the waiting time of long/short transactions by using probabilistic method. Furthermore, we verified it by simulation study. These results reveal that our method is more effective in terms of short transactions than usual two-phase locking.
|
