R. Bayer, H. Heller, and A. Reiser. Parallelism and recovery in database systems. ACM Trans. on Database Systems, 5(2):139--156, June 1980.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....section, we discuss related work on multi version concurrency control schemes and the relationship of our work with deadline cognizant concurrency control algorithms proposed in the real time database systems literature. A number of versioning schemes have been proposed for disk based databases [BHR80, CFL 82, MPL92, AS89, BC92b] Our logical versioning scheme is tailored for main memory systems since it eliminates storage space overheads for items with a single version and it garbage collects old versions aggressively. We also present the actions performed on indices when items are ....

R. Bayer, H. Heller, and A. Reiser. Parallelism and recovery in database systems. ACM Trans. on Database Systems, 5(2):139--156, June 1980.

....for this purpose. 3.3 Two Phase COMMIT Protocol: Single Logging Solution In the following paragraphs, we will concentrate on the realization of REDO logging. Nevertheless, we have to provide a strategy for UNDO logging. Because of the use of an RX synchronization protocol (instead of RAX [1]) we have to perform ROLLBACK on the owner s page copy in case of ABORT, in order to reconstruct the previous consistent page version, which is not necessarily in the physical database. To ensure the WAL principle, the page owner (PO) must log the BFIM of the local page copy before admitting ....

....information and REDO recovery on each node, as it was necessary in the first case, when PO recovery required the global lock information to be in a consistent state before it could start execution. PO node 3 R GLM FIGURE 5. Example: Reconstruction of the Owner Table with i am PO Messages ver. [2, 1 ] page 15 cps isowner T 15, 3, 2, 1 ] 17, 1, 1, 3 ] 21, 2, 0, 5 ] PO node 2 R 2 ver. 0, 5 ] page 21 cps isowner T PO node 1 ver. 1, 3 ] page 17 cps 5, 7 isowner T Log 2 Log 3 i am PO (21, 0, 5 ] i am PO (17, 1, 3 ] 5 Related Work ARIES In [11] ....

[Article contains additional citation context not shown here]

Bayer R., Heller H., Reiser A. Parallelism and Recovery in Database Systems. In ACM Transactions on Database Systems, Vol. 5, No. 2, June 1980, pp. 139 - 156. -

....The picture is somewhat different for distributed database CCMs where comparatively, very few performance studies have been done. There is an increasing interest in the performance of distributed database systems, in particular the use of multiple versions of data for improving concurrency [1, 2, 3, 4, 5, 6, 15, 16]. The basic idea projected in these papers is to support more than one version of a data item and to allow a transaction to access the correct version. Informally, a correct version of a data item is the version, which a transaction would have used if it would have run serially. For example, ....

....which created x i committed successfully. In a single version approach, T 1 would have either rolled back or blocked depending upon the underlying CCM. The multiversion approach has a number of incarnations, each proposing a different method for using a version. For example, the scheme proposed in [1, 6] maintains only one version of a data item whereas in [2, 3, 4, 5] a number of versions are maintained. The algorithms that support multiple versions of a data item guarantee that read only transactions will not be rolled back and a write write conflict will not arise. This means that read only ....

Bayer, R., Heller, H., and Reiser, A. Parallelism and recovery in database systems. ACM Trans. Database Syst. 5, 2 (June 1980), 139-156.

....example, the work of [3] relies on a notion of recoverability that allows a transaction to commit after reading non committed data. This data can be fixed in case of abort of its producing transaction. A scheme that maintains for each object an old value and a new value has been proposed in [4]. So, transactions may utilize the old value if the new one is being updated. The scheme increases parallelism. A locking scheme is provided to implement the idea with some unwanted side effects, e.g. a transaction may obtain the old value, for consistency reasons, even after a new committed value ....

R. Bayer, H. Heller, and A. Reiser, "Parallelism and Recovery in Database Systems," ACM Trans. on Database Systems, 5(2), June 1980, pp. 139-156.

....transaction consistent, version [8, 9, 7, 10, 11, 12, 13, 14] Depending on the total number of versions maintained for each data item, most existing multiversioning schemes can be further classified into two categories: two versions and an unrestricted number of versions. Two version approaches [15, 16] reduce, but do not eliminate, data contention between transactions and queries because only one old version is maintained. Thus, contention still exists. On the other hand, at the expense of high storage overhead and complex version management, approaches using an unrestricted number of versions ....

...., QL 1 , and QL 0 need to be changed. As a result, the impact of a querysnapshot advancement on delaying transactions from being committed is negligible. 5 Related work 5. 1 Two versions Bayer et al. and Stearns and Rosenkrantz have proposed multiversion concurrency protocols using two versions [15, 16]. Data contention is reduced but not eliminated. The increase in the level of concurrency is limited because only one single old version is maintained, and queries may still compete with transactions through special locking protocols. 5.2 Unrestricted number of versions To serialize a query ....

R. Bayer, H. Heller, and A. Reiser, "Parallelism and recovery in database systems," ACM Trans. on Database Systems, vol. 5, no. 2, pp. 139--156, June 1980.

.... issued by read only queries must update timestamp information that is physically associated with each object (possibly turning reads into writes) Several 2PL based algorithms that retain at most two versions of data in order to reduce blocking due to read write conflicts have also been proposed [Baye80, Stea81]. Again, however, these schemes are not ideallysuited for use with long running read only queries. They either allow update transactions to abort queries, or they allow queries to block or abort update transactions, neither of which is particularly desirable in an OLTP setting. ....

Bayer, et al., "Parallelism and Recovery in Database Systems," ACM Trans. on Database Sys., 5(2), June 1980.

....transactions and queries can be reduced, or eliminated: transactions create a new version of an object upon each update, while queries access an old, but consistent, version. Various multiversioning schemes have been proposed to increase the level of concurrency between transactions and queries [2, 3, 4, 7, 19]. However, the increases in storage overhead and version management complexity, including garbage collection, may become significant [1, 8, 9, 17, 20] Recently, different schemes have been proposed to address these problems by maintaining a fixed number of database snapshots [5, 15, 21] A ....

R. Bayer, H. Heller, and A. Reiser. Parallelism and recovery in database systems. ACM Trans. on Database Systems, 5(2):139--156, June 1980.

....of our knowledge, Reed s distributed timestamp ordering scheme [Reed83] was actually the first multiversion concurrency control algorithm proposal. Several 2PL based algorithms that retain at most two versions of data in order to reduce blocking due to read write conflicts have also been proposed [5, 27]. Other multiversion extensions of single version concurrency control algorithms include: multiversion optimistic validation [25, 9, 19] multiversion timestamp ordering [2] and the multiversion tree protocol [26] Finally, in [7] we presented a generalization of MV2PL that provides queries ....

Bayer, et al., "Parallelism and Recovery in Database Systems," ACM Trans. on Database Sys., 5(2), June 1980.

....VDBC assumes a strict two phase locking protocol. In the simulation studies described in section 3 we use the primary copy locking algorithm [13] which corresponds to distributed locking with fixed global lock authorities. The locking protocol used at the global lock authorities is known as RAX [1]. The allocation of the lock authorities is done with the same hash function we use for the page administrator. In this case, lock messages and page owner request messages are combined, which leads to reduction of communication cost. 3. Performance In this section we present the results of the ....

Bayer, R., Heller, H. Reiser, A.: Parallelism and Recovery in Database Systems, in: ACM TODS, 5 (2), 1980.

....an attempt is made to reclaim that version s space. 7 Related Work In this section, we discuss related work on main memory databases, multi version concurrency control schemes and concurrency control schemes for indices. A number of versioning schemes have been proposed for disk based databases [BHR80, CFL 82, MPL92, AS89, BC92b] Our logical versioning scheme is tailored for main memory systems since it eliminates storage space overheads for items with a single version and it garbage collects old versions aggressively. We also present the actions performed on indices when items are ....

R. Bayer, H. Heller, and A. Reiser. Parallelism and recovery in database systems. ACM Trans. on Database Systems, 5(2):139--156, June 1980.

....access to the data. 1 Introduction Databases often maintain multiple versions of data to avoid interference between read only queries and update transactions. Many multi versioning concurrency control protocols have been proposed, both for centralized and distributed databases (see, e.g. SR81, BHR80, CG85, Wei87, CFL 82, MPL92, AS89, BC92] Of special interest to us are protocols that satisfy the following three properties (called the non interference requirement) 1) completely decouple read only queries from updates, 2) never delay transactions on account of version management, and ....

....much heavier weight updates) At the same time, while read only queries should not fall too far behind the updates, it is not critical that the queries read the latest committed data. This non interference requirement eliminates from consideration many protocols, like the twoversion protocols of [BHR80, SR81] where interference between read and update transactions is still possible, and the protocol of [Wei87] that requires read transactions to write read timestamps to data items they access. Existing techniques that satisfy the non interference requirement fall into two categories. Techniques ....

[Article contains additional citation context not shown here]

R. Bayer, H. Heller, and A. Reiser. Parallelism and recovery in database systems. ACM Trans. on Database Systems, 5(2):139--156, June 1980.

No context found.

R. Bayer, H. Heller, and A. Reiser. Parallelism and recovery in database systems. ACM Trans. on Database Systems, 5(2):139--156, June 1980.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC