C. Mohan, H. Pirahesh, and R. Lorie. Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions. In Proc. of ACM SIGMOD Int. Conf. on Management of Data, pages 124--133, 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....traditional multiversion concurrency control algorithms [3] the reader will read an older version than the version that the writer is currently writing to, thus avoiding the conflicts. In particular, we further differentiate between two categories of multiversion algorithms, i.e. finite version [15, 11] and unrestricted version [2] algorithms. Finite version algorithms pre set a fixed number of maximal versions (some constant n) There is a so called version expiration [15, 11] phase when the number of versions exceeds the allowed maximum. In that case, we have to switch the reader to a newer ....

....In particular, we further differentiate between two categories of multiversion algorithms, i.e. finite version [15, 11] and unrestricted version [2] algorithms. Finite version algorithms pre set a fixed number of maximal versions (some constant n) There is a so called version expiration [15, 11] phase when the number of versions exceeds the allowed maximum. In that case, we have to switch the reader to a newer existing version to allow for the writers to start making a new version. To achieve this, we require that either the writer synchronizes with all readers on the old version or ....

C. Mohan, H.Pirahesh, and R. Lorie. Efficient and Flexible Methods for Transient Versioning of Records to Avoid Locking by Read-only Transactions. In Proceedings of SIGMOD, pages 124-133, 1992.

....a read only transaction may still have to wait on locks held by an update transaction, which may in turn be waiting on a different transaction, or on disk writes to the log. These waits become a serious source of unpredictability for response times. Multiversion concurrency control methods [MPL92, Had88, AS89, BG83, BC92a, IKK90, CFL 82] prevent update transactions from conflicting with read only transactions by providing the latter with a consistent but somewhat out of date view of the database. In order to provide this view, multiple versions of recently updated data items are ....

....but somewhat out of date view of the database. In order to provide this view, multiple versions of recently updated data items are retained. Early multi version schemes [Ree78] used timestamps for readers and writers, but more recent multi version locking schemes [CFL 82, AS89, BC92a, MPL92] use timestamps with read only transactions, allowing them to use old versions without locking, while requiring updaters to perform locking. However, none of the above techniques guarantees complete isolation of read only transactions from update transactions in a system, since the access path to ....

[Article contains additional citation context not shown here]

C. Mohan, H. Pirahesh, and R. Lorte. Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions. In ACM SIGMOD Conf. on the Management of Data 92, San Diego., June 1992.

....how frequently changes are made to the source tables and the need for the reader sessions to access the latest data. Thus, the crux of the matter is to determine, 1. how to execute the reader sessions and the periodic maintenance transaction on the view concurrently without blocking each other [BG83,MPL92]. 2. how to maintain session consistency. The reader session can have more than one queries on the same view. The results of those queries should always be consistent. Suppose, an analyst wanted to find the total sales made by stores in each city. A common subsequent action for the analyst would ....

C..Mohan, H.Pirahesh, and R.Lorie. Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions. In Proceedings of ACM SIGMOD 1992 International Conference on Management of Data, pages 124-133,1992.

....of read and update operations of competing transactions, either by restricting access to data or by replicating the data. This section presents an overview of mechanisms that guarantee repeatable read isolation by restricting data access; for approaches that employ replication, see for example [MPL92, Stu95] The simplest solution would be to lock all involved tables for the duration of the entire transaction. Unfortunately, this leads to an unacceptably low degree of concurrency. DBMSs try to avoid this by accessing the tables through index structures and explicitly locking only as much as ....

C. Mohan, H. Pirahesh, and R. Lorie. Efficient and Flexible Methods for Transient Versioning of Records to Avoid Locking by Read-Only Transactions. In Proceedings of the ACM-SIGMOD International Conference on Management of Data, San Diego, California, June 1992.

....all of the desirable work between the time of the backup and the time of recovery is lost. Keeping this window of vulnerability acceptably low incurs a substantial cost in maintaining frequent backups or checkpoints, although there are algorithms for efficiently establishing snapshots on the fly [AJM95, MPL92, Pu86]. The compensation approach [GM83, GMS87] seeks to undo either committed transactions or committed steps in long duration or nested transactions [KLS90]without necessarily restoring the data state to appear as if the malicious transactions or steps had never executed. There are two kinds of ....

C. Mohan, H. Pirahesh, and R. Lorie. Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions. In Proceedings of ACM SIGMOD International Conference on Management of Data, pages 124--133, San Diego, CA, June 1992.

....value for each data item, maintains and broadcasts multiple versions per item. Multiversion schemes, where older copies of items are kept for concurrency control purposes, have been successfully used to speed up processing of online read only transactions in traditional pull based systems (e.g. [13]) Let c 0 be the bcycle during which a client transaction R performs its first read operation. During c 0 , transaction R reads the most up to date value for each data item, that is, the value having the largest version number. In later cycles, R reads the value with the largest version number ....

C. Mohan, H. Pirahesh, and R. Lorie. Efficient and Flexible Methods for Transient Versioning to Avoid Locking by Read-Only Transactions. In Proc. of SIGMOD, 1992.

....of the desirable work between the time of the 5 backup and the time of recovery is lost. Keeping this window of vulnerability acceptably low incurs a substantial cost in maintaining frequent backups or checkpoints, although there are algorithms for efficiently establishing snapshots on the fly [AJM95, MPL92, Pu86]. The compensation approach [GM83, GMS87] seeks to undo either committed transactions or committed steps in long duration or nested transactions [KLS90]without necessarily restoring the data state to appear as if the malicious transactions or steps had never executed. There are two kinds of ....

C. Mohan, H. Pirahesh, and R. Lorie. Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions. In Proceedings of ACM SIGMOD International Conference on Management of Data, pages 124--133, San Diego, CA, June 1992.

....all of the desirable work between the time of the backup and the time of recovery is lost. Keeping this window of vulnerability acceptably low incurs a substantial cost in maintaining frequent backups or checkpoints, although there are algorithms for efficiently establishing snapshots on the fly [2, 18, 19]. The compensation approach [6, 7] seeks to undo either committed transactions or committed steps in long duration or nested transactions [9] without necessarily restoring the data state to appear as if the malicious transactions or steps had never executed. There are two kinds of compensation: ....

C. Mohan, H. Pirahesh, and R. Lorie. Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions. In Proceedings of ACM SIGMOD International Conference on Management of Data, pages 124--133, San Diego, CA, June 1992.

....particularly in the presence of the UF1 and UF2 update functions in the TPC D benchmark. These are important functions to consider in the design of an Active Disk database system, but there are a number of possible methods to minimize the impact of the updates on the decision support queries [Mohan92, Merchant92] and these issues are mentioned again in Section 7.5.2. The discussion that follows will assume that the relations being processed at the disks are uncacheable at the hosts, and that cursor stability is sufficient for the decision support queries. 1 Figure 4 10 shows the entire plan for Query 1 ....

....growth in capacity and the increase in transfer rates) a tradeoff can be made to use additional space in order to efficiently support more sophisticated concurrency control schemes. If space is cheap, an approach such as transient versioning can trade off increased disk usage against performance [Mohan92, Merchant92]. Such a scheme works by using version vectors at the disks to maintain multiple versions of the same page. The primary problem with this approach is the extra disk storage required for the copies of recently modified pages. Given today s drive capacities, this should be a feasible option. The ....

[Article contains additional citation context not shown here]

Mohan, C., Pirahesh, H. and Lorie, R. "Efficient and Flexible Methods for Transient Versioning of Records to Avoid Locking by Read-Only Transactions" SIGMOD, June 1992.

....of a data object that are slightly different in age or in precision are often acceptable. This observation underlies the concept of similarity among data values. A large body of literature exists concerning various approaches to effectively supporting concurrent transaction and query processing [1, 54, 73, 85, 55, 18, 21, 81, 20]. These approaches typically employ multiple versions of data to eliminate read write conflicts between update transactions and read only queries. Queries access transactionconsistent, but maybe out of date, database states. In contrast, ESR does not require multiple versions of data, thus does ....

C. Mohan, H. Pirahesh, and L. Raymond. Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions. In Proceedings of the 1992 ACM SIGMOD International Conference on Management of Data, pages 124--133, San Diego, June 1992.

....may shift depending on which fields are updated. The technique here is to create a new entry (based on the updated value) without actually deleting the previous entry. If the transaction commits, the old entry is deleted. If the transaction aborts, the new entry is removed. As pointed out in [MoPL 92] this approach avoids the next key locking during key delete. Since index entries of deleted keys are maintained atleast until the end of transaction, logical key deletion capability comes automatically. The LU Logging method seems to be suitable for multi processor data base environments. We ....

Mohan, c., Pirahesh, H., Lorie, R. Efficient and Flexible Methods for Transient Versioning of Records to Avoid Locking by Read-Only Transactions, ACM SIGMOD, June 1992.

....writer transaction. Deleting the previous versions causes readers to delay writers because the writer cannot commit until all readers that have read previous versions of modified tuples have committed (otherwise repeatable reads would be sacrificed) MV2PL algorithms [AS89,BC92a,BC92b,CFL 82,MPL92,Wei87,WYC93] guarantee that readers and writers never block each other. 1 MV2PL algorithms maintain enough previous versions of each tuple (or 1 Not all multi version algorithms are MV2PL algorithms. That is, not all use two phase locking to synchronize writers. But since synchronizing writers ....

....readers read the latest version of the tuple that is less than the reader s begin timestamp [CFL 82] This approach requires maintaining possibly many previous versions of each tuple. By carefully choosing the previous tuple versions that are made available to readers, other MV2PL algorithms [MPL92,WYC93] guarantee that readers and writers never block each other with a maximum of three or four tuple versions. All MV2PL algorithms require overhead to access and maintain previous tuple versions. For example, the approach in [CFL 82] stores previous versions in a special version pool on ....

[Article contains additional citation context not shown here]

C. Mohan, H. Pirahesh, and R. Lorie. Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions. In Proceedings of ACM SIGMOD 1992 International Conference on Management of Data, pages 124--133, 1992.

....are both supported at the page level. We chose page level versioning to simplify the implementation of the simulator and to reduce the length of simulation runs. The basic MV2PL and MVQL algorithms are compatible with record level versioning; schemes for record level versioning are discussed in [Bobe92, Moha92]. The version manager divides the database into two segments: the main segment, containing the current versions of pages, and the version pool, containing prior page versions. This organization is similar to the one described in [Chan82] except that we arrange the version pool as a heap of disk ....

Mohan, C., H. Pirahesh, and R. Lorie, "Efficient and Flexible Methods for Transient Versioning of Records to Avoid Locking by Read-Only Transactions," Proc. 1992 ACM SIGMOD Conf., 1992.

....successfully processed, we propose maintaining multiple versions of data items. Multiversion schemes, where older copies of items are kept for concurrency control purposes, have been successfully used to speed up processing of on line read only transactions in traditional pull based systems (e.g. [17]) 3.1 The Basic Multiversion Schemes The basic idea underlying multiversioning is to temporarily retain older versions of data items, so that the number of aborted read only transactions is reduced. Versions correspond to different values at the beginning of each currency interval and version ....

C. Mohan, H. Pirahesh, and R. Lorie. Efficient and Flexible Methods for Transient Versioning to Avoid Locking by Read-Only Transactions. In Proc. of the ACM SIGMOD Intl. Conf. on Management of Data, pp. 124--133, 1992.

....number of versions can eliminate data contention. Recently, various versioning schemes maintaining only a fixed number of versions have been independently proposed to address the problems of storage overhead and version management complexity associ ated with an unrestricted number of versions [10, 17, 18, 19, 20]. In this paper, we present a general class of such fixed versioning schemes, referred to as dynamic finite versioning (DFV) to effectively support concurrent transaction and query processing [20] A detailed comparison and contrast among these different schemes is provided in Section 5.3. DFV ....

....Read only transactions undergo no concurrency control, but the version control mechanism may maintain an unrestricted number of versions for a data item. 5. 3 Fixed number of versions Recently, various implementations which employ only a fixed number of versions have been independently proposed [20, 10, 18, 19]. Bober and Carey proposed an extension to the ring buffer approach developed in [11, 12] by using record level versioning and on page caching for storing prior versions [10] They also introduced the concept of view sharing, which groups together queries to run against the same ....

[Article contains additional citation context not shown here]

C. Mohan, H. Pirahesh, and R. Lorie, "Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions," in Proc. of ACM SIGMOD Int. Conf. on Management of Data, pp. 124--133, 1992.

....E, Simon, D. Tombroff 6 Related Work An extensive literature addresses the problem of designing concurrency control algorithms that augment the performance of concurrent transactions. Our work directly builds on previous work on MV2PL protocols ( CFL 82] BHG87] AS89] AK91] BC92a] MPL92] but differs from those by focusing on the specific class of writethen read transactions. In [BC92a] and [MPL92] the authors propose techniques in which an update transaction does not systematically create a new version of its updated items. Hence, read only transactions do not access ....

....algorithms that augment the performance of concurrent transactions. Our work directly builds on previous work on MV2PL protocols ( CFL 82] BHG87] AS89] AK91] BC92a] MPL92] but differs from those by focusing on the specific class of writethen read transactions. In [BC92a] and [MPL92] the authors propose techniques in which an update transaction does not systematically create a new version of its updated items. Hence, read only transactions do not access versions which are the most up to date before their starting time. Instead, they all read a given older state which is ....

C. Mohan, H. Pirahesh, and R. Lorie. Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions. Proc. ACM SIGMOD Int. Conf. on Management of Data, San Diego, California, pages 124--133, June 1992.

....have been garbage collected prematurely. Thus, both storage overhead and version management complexity, such as version selection and garbage collection, can be a severe problem. Recently, a new class of mvCC algorithms, such as dynamic finite versioning (DFV) 16, 8] and transient versioning [9], have been proposed to allow a choice in the trade off between the number of versions available (i.e. the storage space occupied) and the obsolescence of data being accessed. Queries are guaranteed to read from a consistent database state, but that consistent state may be out of date [8] ....

C. Mohan, H. Pirahesh, and R. Lorie. Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions. In Proc. of ACM SIGMOD Int. Conf. on Management of Data, pages 124--133, 1992.

....and wait for all currently executing updaters to finish before entering the system. While the query is waiting, subsequently arriving update transactions would not be allowed to discard their prior versions before commit. 2 Similar ideas, developed independently, are presented in [Wu91] and [Moha92]. 16 same startup timestamp, t b , or it may decide to generate a new logical view and run with a startup timestamp of t d . After the transaction begins, a transaction that has updated X (generating X 2 ) subsequently commits with a timestamp of t e . If Q 2 decides to share Q 1 s startup ....

....Lehm81, Moha90] Because (non 2PL) B tree concurrency control algorithms are widely viewed as being important to achieving acceptable performance, we do not consider the Rdb approach further. A number of other multiversion indexing approaches have been proposed in the literature; examples include [East86, Ston87, Kolo89, Lome89, Lome90, Moha92]. With the exception of [Moha92] however, all of these ################################ 1 Using an index only plan, a query computing the average salary of a group of employees, for example, does not have to retrieve the employee tuples if an index on employee salary exists; instead it can ....

[Article contains additional citation context not shown here]

Mohan, C., H. Pirahesh, and R. Lorie, "Efficient and Flexible Methods for Transient Versioning of Records to Avoid Locking by Read-Only Transactions," Proc. 1992 ACM SIGMOD Conf., 1992.

....of transactions. 5 Related Work An extensive literature addresses the problem of designing concurrency control algorithms that augment the performance of concurrent transactions. Our work directly builds on previous work on MV2PL protocols ( CFL 82] BHG87] AS89] AK91] BC92] MPL92] but differs from those by focusing on the problem of optimizing read operations in update transactions. The 2V2PL multiversion protocol authorizes the use of versions in update transactions, and guarantees serializability of transactions. In this protocol, there are three modes of locks ....

C. Mohan, H. Pirahesh, and R. Lorie. Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions. Proc. ACM SIGMOD Int. Conf. on Management of Data, San Diego, California, pages 124--133, June 1992.

....do not contribute to data contention since they do not have to set or wait for locks. This form of versioning, where old copies of data are retained temporarily for concurrency control purposes (as opposed to long term retention for historical queries) has been referred to as transient versioning [22]. Since indexes are important for good performance in database systems, it is important to to determine how they may coexist with MV2PL. Conventional single version indexing structures such as B trees and hashing are not entirely compatible with MV2PL in their current forms, as they support ....

.... Rdb system, is to treat index nodes like data records at the storage level, including having MV2PL applied to them [17] While this approach supports index only plans, it is not compatible with the use of high performance non 2PL B tree concurrency control algorithms such as those proposed in [3, 21, 22]. Because (non 2PL) B tree concurrency control algorithms are widely viewed as being important to achieving acceptable performance, we do not consider the Rdb approach further. A number of other multiversion indexing approaches have been proposed in the literature; examples include [13, 28, 18, ....

[Article contains additional citation context not shown here]

Mohan, C., H. Pirahesh, and R. Lorie, "Efficient and Flexible Methods for Transient Versioning of Records to Avoid Locking by Read-Only Transactions," Proc. 1992 ACM SIGMOD Conf., 1992.

....will be a small fraction of the execution time even for much longer transactions than those used in these simulation experiments. 5 Related Work Transient versioning is an alternative approach to avoid lock contention while still achieving transactionconsistency for queries in an OLTP system [5, 3, 12, 19]. In transient versioning algorithms, transactions create a new physical version of a data item when performing an update. Queries may access an older version in order to get a transaction consistent view. In order to maintain data clustering, transient versioning algorithms should perform ....

C. Mohan, H. Pirahesh, and R. Lorie. Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions. In Proc. ACM SIGMOD, pp. 124--133, June 1992.

No context found.

C. Mohan, H. Pirahesh, and R. Lorie. Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions. In Proc. of ACM SIGMOD Int. Conf. on Management of Data, pages 124--133, 1992.

No context found.

C. Mohan, H. Pirahesh, and R. Lorie, "Efficient and Flexible Methods for Transient Versioning to Avoid Locking by Read-Only Transactions," Proc. ACM SIGMOD Int'l Conf. Management of Data, pp. 124-133, 1992.

No context found.

C. Mohan, H. Pirahesh, and R. Lorie, "Efficient and Flexible Methods for Transient Versioning to Avoid Locking by Read-Only Transactions," Proc. ACM SIGMOD Int'l Conf. Management of Data, pp. 124-133, 1992.

No context found.

C. Mohan, H. Pirahesh, and R. Lorie. Efficient and flexible methods for transient versioning of records to avoid locking by read-only transactions. In Proceedings of the ACM SIGMOD International Conference on Management of Data, pages 124--133, San Diego, CA, June 1992.

First 50 documents

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