Hybrid concurrency control and recovery for multi-level transactions
Multi-level transaction schedulers adapt confiict-serializability on different levels. They exploit the fact that many low-level conflicts (e.g. on the level of pages) become irrelevant, if higher-level application semantics is taken into account. Multi-level transactions may lead to an increase in...
Elmentve itt :
| Szerzők: | |
|---|---|
| Dokumentumtípus: | Cikk |
| Megjelent: |
2000
|
| Sorozat: | Acta cybernetica
14 No. 3 |
| Kulcsszavak: | Számítástechnika, Kibernetika |
| Tárgyszavak: | |
| Online Access: | http://acta.bibl.u-szeged.hu/12636 |
| Tartalmi kivonat: | Multi-level transaction schedulers adapt confiict-serializability on different levels. They exploit the fact that many low-level conflicts (e.g. on the level of pages) become irrelevant, if higher-level application semantics is taken into account. Multi-level transactions may lead to an increase in concurrency. It is easy to generalize locking protocols to the case of multi-level transactions. In this, however, the possibility of deadlocks may diminish the increase in concurrency. This stimulates the investigation of optimistic or hybrid approaches to concurrency control. Until now no hybrid concurrency control protocol for multi-level transactions has been published. The new FoPL protocol (Forward oriented Concurrency Control with Preordered Locking) is such a protocol. It employs access lists on the database objects and forward oriented commit validation. The basic test on all levels is based on the reordering of the access lists. When combined with queueing and deadlock detection, the protocol is not only sound, but also complete for multi-level serializable schedules. This is definitely an advantage of FoPL compared with locking protocols. The complexity of deadlock detection is not crucial, since waiting transactions do not hold locks on database objects. Furthermore, the basic FoPL protocol can be optimized in various ways. Since the concurrency control protocol may force transactions to be aborted, it is necessary to support operation logging. It is shown that as well as multi-level locking protocols can be easily coupled with the ARIES algorithms. This also solves the problem of rollback during normal processing and crash recovery. |
|---|---|
| Terjedelem/Fizikai jellemzők: | 419-453 |
| ISSN: | 0324-721X |