David J. DeWitt, David Maier, Philippe Futtersack, and Fernando Velez. A Study of Three Alternative Workstation-Server Architectures for Object-Oriented Database Systems. In Proceedings of the 16th International Conference on Very Large Databases, Brisbane, Australia (VLDB '90), pages 107--121, August 1990.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....order of benefit while counting the new tuples accessed by a query to represent its replaceable size. We are evaluating this and other approaches in our ongoing work. 4 Related work Client server database systems have also addressed replacement issues in page based and tuple based client caches [5, 6]. Replacement in a cache indexed by semantic units such as views or query results introduces different challenges, however. Semantic caches have been proposed in client server database systems [3] but that work addressed only read only caching and used a different storage implementation. ....

D. DeWitt, P. Futtersack, D. Maier, and F. Velez. A study of three alternative workstation-server architectures for object-oriented database systems. In VLDB Conference, pages 107--121, 1990.

....try to exploit spatial locality assuming that clustering of tuples to pages is effective. Page level caching assumes that only pages from the base relations are stored in the cache. Lots of work has been done in page level caching, FCL92] FCL93] FC94] Fra96] FCL97] etc. In the work of [DM90], the caching mechanism with a granularity of a single object is implemented and compared with a page level caching. CSL98] investigates three different levels of granularity of caching, namely, attribute caching which caches frequently accessed attributes of database objects, object caching ....

David J. DeWitt and David Mater, A Study of Three Alternative Workstation-Server Architectures for Object-Oriented Database Systems, Proceedings of the VLDB Conference, 1990 28

....the more traditional page based architecture where all interaction between clients and servers takes place at the granularity of individual pages [7, 11, 12, 1] Other systems [9, 5] use object server architectures but do not specifically address the problem of installation reads. Dewitt et al. [6] is one of the first studies that investigated the design choices for a persistent object system architecture. The study focused on the question of distributing the functionality of the persistent object system between the client and the server. It measured and compared a page based system and an ....

David J. DeWitt, Philippe Futtersack, David Maier, and Fernando Velez. A study of three alternative workstation-server architectures for object oriented database systems. In Proceedings of the 16th Conference on Very Large Data Bases, pages 107--121, Brisbane, Australia, 1990.

....and copy requested or prefetched object versions into the object cache, the object cache s task is to efficiently maintain those object versions, i.e. it is used as data storage memory. Note that our intuition about this cache structure has been confumed experimentally by a performance study [DMF 90] demonstrating that an object based caching architecture is superior to a page based caching architecture when physical clustering is poor and the client s cache size is small relative to the size of the database which is typical for a mobile environment. We further assume that the broadcast ....

D. J. DeWitt, D. Maier, P. Futtersack, and F. Velez. A study of three alternative workstation-server architectures for object-oriented database systems. VLDB 1990, pages 107-121.

....into the main memory before the user explicitly speci es her next request. 3 Semantic Caching for Visualization Applications Semantic caching is a popular caching strategy proposed in recent years [24, 8] for providing ecient support for access to data. In contrast to traditional caching schemes [10, 25], it caches query descriptors rather than pages of data or individual objects. It provides the following bene ts over the traditional syntactic caching approaches: adjusts grouping of queries to the requirements of the incoming query so that no irrelevant data is cached along with the relevant ....

D. DeWitt, D. Mayer, P. Futtersack, and F. Velez. A study of three alternative workstation-server architectures for object-oriented database systems. In Proc of VLDB'90 16th Intl Conf on Very Large Data Bases, Brisbane, Australia,

.... processing (e.g. through the materialization of object references) Such an approach is especially required in the scope of client server environments (also called workstation server environments) to minimize communication traffic between different system components located on separate machines [7] [14] 18] Since the expressive power of the query language should be available not only for loading unloading the buffer (i.e. check out check in operations) but also for knowledge processing tasks within the application (in the scope of the buffer) main memory based query processing has to ....

DeWitt, D.J., Futtersack, P., Maier, D., Velez, F.: A Study of Three Alternative Workstation Server Architectures for Object-Oriented Database Systems, Proc. 16th VLDB Conf., Brisbane, Australia, 1990.

....2 P. Godfrey J. Gryz In distributed database environments, another layer of caching is possible: to cache information between servers and clients. Two caching approaches, page caching and tuple caching in which memory pages or tuples are cached, respectively, have been studied in this context [4, 10]. Semantic query caching (SQC) offers a third approach. In [4] it is shown that semantic caching may outperform the page and tuple caching due to semantic locality: subsequent queries often are related conceptually with previous queries, so ultimately will be pulling data from the same logical ....

D. DeWitt, P. Futtersack, D. Maier, and F. Velez. A study of three alternative workstation-server architectures for object-oriented database systems. In Proceedings of VLDB, 1990.

....should be at least two ways to specify collections, one using enumeration of members and one using the query language to specify membership. The OODB literature suggests specifying sets by enumerating the members of a set, typically by means of a linked list or array of identifiers for members [DEWI90]. We believe that this specification is generally an inferior choice. To explore our reasoning, consider the following example. ALUMNI (name, age, address) GROUPS (g name, composition) Here we have a collection of alumni for a particular university along with a collection of groups of alumni. ....

Dewitt, D. et. al., "A Study of Three Alternative Workstation-Server Architectures for Object Oriented Database Systems," ALTAIR Technical Report 42-90, Le Chesnay, France, January 1990.

.... respective fields (OLTP and Decision Support) TPC E failed to garner enough support because being an enterprise benchmark, it was only relevant to a relatively small number of companies competing in that space [11] The Simple Database Operations Benchmark [9] Altair Complex Object Benchmark [14], OO1 [6] and OO7 [3, 2] Benchmarks were created to provide useful insight for end users evaluating the performance of Object Oriented Database Management Systems (OODBMS) Most object relational DBMS are build on top of relational database by adding the following four key features [12] ....

D. DeWitt, P. Futtersack, D. Maier and F. Velez. A study of three alternative workstation-server architectures for object oriented database systems. Proceedings of the Sixteenth Very Large Data Bases Conference, Brisbane, Australia, 1990. pp. 107-121.

....lead to better overall system performance and scalability. Object oriented database systems (OODBs) have gained much popularity in recent years, especially in the application areas of CAD CAM and CASE. OODBs typically have client server architectures, and commonly use the data shipping approach [5] to move relevant data from the central database to local client caches. The data shipping strategy allows query processing to occur locally at client sites, and exploits the resources of client workstations. A page (of 4K or 8K fixed byte length) is usually the physical unit of data transfer ....

....closely together in some set of data pages. In contrast, a range predicate on an unclustered B tree index will first access a consecutive set of index page entries, but they will reference objects in random data pages. Thus, the number of data objects accessed per page, i.e. the page locality [5], may differ radically depending on whether the data is accessed via a clustered or an unclustered index. The simulation model of [3] does have page locality as a parameter, but index access and update costs are not considered. ffl Index write probability: Previous simulation studies have ....

[Article contains additional citation context not shown here]

D. J. DeWitt, D. Maier, P. Futtersack, and F. Velez, "A Study of Three Alternative Workstation-Server Architectures for ObjectOriented Database Systems," 16th Int. Conf. on Very Large Data Bases, Brisbane, Australia, 1990, pp. 107-121.

....and client server EXODUS [CDG 90] employ the data shipping architecture. They use a page as the unit of transfer between the client and the server. This is called the page server architecture. Performance tradeoffs between page and object shipping architectures are examined by DeWitt et al. DFMV90] The page server architecture is employed by most object database servers. The pages requested by a client are transferred to the client workstations where they are processed. Processing data at the client makes use of the computing resources at the client workstations. In a multi server ....

D. DeWitt, P. Futtersack, D. Maier, and F. Velez. A Study of Three Alternative Workstation Server Architectures for Object Oriented Database SYstems. In Proceedings of the 16th International Conference on Very Large Data Bases, Aug 1990.

....for preserving the integrity of the data and for enforcing transaction semantics. Data shipping systems can be structured as page servers, which send physical units of data between servers and clients, and object servers, in which clients and servers interact using logical units of data [DeWitt 1990, Carey 94] 2 Each Client DBMS process is responsible for translating local application data requests into requests for specific database items (i.e. pages or objects) and for bringing those items into memory at the client. As a result, all of the data items required by an application are ....

D. DeWitt, P. Futtersack, D. Maier, and F. Velez. A Study of Three Alternative Workstation-Server Architectures for Object-Oriented Database Systems. In Proc. of the 16th VLDB Int. Conf., Brisbane, Australia, August 1990.

....Reserved. 3 GemStone, along with Vbase [Andre90] was one of the first ODBMSs commercially available. GemStone is based on the Smalltalk object oriented language with extensions to support DBMS services. It has a client server architecture (Figure 1) and can be described as using a page server [DeWit90]. It can also be categorised as an active 1 ODBMS with the ability to execute methods on the server. This feature can be very useful in querying large collections of objects on the server and transferring only the results back to the client, rather than shipping large numbers of objects from the ....

....large quantities of data [Butte91] 2.2 O 2 System The O 2 ODBMS was originally developed as a research prototype and was subsequently released as a commercial product. Originally, it was based on an object server architecture, but because cache inconsistencies were discovered and reported in [DeWit90], the architecture was subsequently changed to a page server. O 2 can be classed as an active ODBMS, since indexed access for large collections is possible on the server [Manol94] Furthermore, the Object Query Language (OQL) also executes on the server. Copyright 1996 A.B. Chaudhri. All Rights ....

[Article contains additional citation context not shown here]

D. DeWitt, P. Futtersack, D. Maier and F. Velez (1990) A study of three alternative workstation-server architectures for object oriented database systems. Proceedings of the Sixteenth Very Large Data Bases Conference, Brisbane, Australia, 1990, pp. 107-121.

....is needed, adding new workstations is sufficient and is easy to do. 2.4 Cache management Caches aim at minimizing the cost of data retrievals. In this section, we describe two aspects that have an impact on the efficiency of the cache: cache integration and cache cooperation. Studies such as [DFMV90] have shown that page server architectures are more efficient than record server architectures. Moreover, as the former are the most used in today s implementation, we only consider caches of pages, each page containing one or several records or just a part of a big record. Cache management is ....

D.J. DeWitt, P. Futtersack, D. Maier, and F. Velez. A study of three alternative workstationserver architectures for object oriented database systems. In Proceedings of the 16th VLDB Conference, Brisbane, Australia, 1990.

....The cooperation between a server and a number of workstations in an engineering design environment is examined in [13] The DBMS prototype that supports a multi level communication between workstations and server which tries to reduce redundant work at both ends is described. DeWitt et al. [8] examine the performance of three workstation server architectures from the Object Oriented DBMS point of view. Three approaches in building a server are proposed: object, server and file server. A detailed simulation study is presented with different loads but no concurrency control. They ....

D. DeWitt, D. Maier, P. Futtersack, and F. Velez. A Study of Three Alternative Workstation-- Server Architectures for Object--Oriented Database Systems. In Proceedings of the 16th Very Large Data Bases Conference, pages 107--121, Brisbane, Australia, 1990.

.... and Their Integration (ADTI 94) Nara, Japan, October 1994 of fast local area networks paved the way to the introduction of the Client Server Database Systems (CS DBMSs) The central concept in CS DBMSs is that a dedicated machine runs a DBMS and maintains a server database (DBMS Server)[7, 8, 5]. The users of the system access the database through either their workstations or PCs via a local area network. They usually run their application programs locally on their own machines and direct all database inquiries and or updates to the DBMS Server. In this way, they become the server s ....

D. DeWitt, D. Maier, P. Futtersack, and F. Velez. A Study of Three Alternative Workstation--Server Architectures for Object-- Oriented Database Systems. In Proceedings of the 16th Very Large Data Bases Conference, pages 107--121, Brisbane, Australia, 1990.

....its own memory space. Subsequently, the query processor running on the local CPU refers to these staged pages to generate the result(s) for the client application query. The usage of client buffer space to hold a portion of the server database has proved to be a basic yet effective form of caching [69, 24, 9, 19]. This caching plays a central role in the improvement of performance rates of the architecture [19] as compared to those achieved by the basic CS configuration. Figure 6 shows the functional components of the architecture in discussion. Shared Database Application Software LAN Client Server ....

....due to the increased locality of access, a large server cache can provide benefits of similar magnitude without the expense of a local disk cache. The three main data shipping classes of CS architectures useful for object oriented databases are the page server, object server and file server [24]. These differ principally in the granularity of data transfer and caching. The file server and page server have their origins in distributed file systems. The following subsections examine each of the above classes in some detail. 3.4.1 File Server CSDs While this method is not really a major ....

[Article contains additional citation context not shown here]

D. DeWitt, D. Maier, P. Futtersack, and F. Velez. A Study of Three Alternative Workstation--Server Architectures for Object--Oriented Database Systems. In Proceedings of the 16th International Conference on Very Large Data Bases, pages 107--121, 1990.

....are designed to operate in a client server environment [6, 13, 14, 22] There is usually a centralized server and multiple clients that are connected to the server by a local area network. The two most common models for database processing in such systems are data shipping and query shipping [8]. In this paper, we deal with the data shipping client server model where the data is shipped from the server to the client to be processed there. This model has become quite pervasive in database computing. However, most previous efforts on this model assume one or more of the following: i) the ....

D. DeWitt, D. Maier, P. Futtersack, and F. Velez. A Study of Three Alternative Workstation--Server Architectures for Object--Oriented Database Systems. In Proceedings of the 16th International Conference on Very Large Data Bases, pages 107--121, Brisbane, Australia, 1990.

....Rubinstein et al. in [RKC87] presented the RAD Unify type of DBMS architecture where the server executes low level database operations (locking and page handling) while diskless workstations perform query processing and use their virtual memory to improve query response time. DeWitt et al. DMFV90] examine the performance of three workstation server architectures from the Object Oriented DBMS point of view. Wilkinson and Niemat in [WN90] propose two concurrency control algorithms for maintaining consistency of workstation cached data. Alonso et al. in [ABGM90] support the idea that ....

D. DeWitt, D. Maier, P. Futtersack, and F. Velez. A Study of Three Alternative Workstation--Server Architectures for Object--Oriented Database Systems. In Proceedings of the 16th Very Large Data Bases Conference, pages 107--121, Brisbane, Australia, 1990.

....collection and hence not designed with garbage collection in mind have already been built into those systems. Three such criteria are: 1) partition objects based on access patterns (e.g. 25] 29] 2) partition objects based on the unit of transfer between a server and a client (e.g. [16] ) and 3) partition objects to increase locking granularity and thereby decrease overhead (e.g. 8] The previous discussion brings to light an important distinction between programming language systems and ODBMSs that makes the design of garbage collection algorithms for ODBMSs especially ....

D.J. Dewitt, P. Futtersack, D. Maier, and F. Velez, "A Study of Three Alternative Workstation-Server Architectures for ObjectOriented Database Systems," Proc. 16th Int'l Conf. Very Large Data Bases, pp. 107--121. Morgan Kaufmann, Aug. 1990.

....hardware and software. In recent years, the Client Server computing paradigm [27] has become a popular vehicle for the development of modern database systems. It provides the necessary infrastructure that potentially may satisfy both the high performance and the scalability requirements[11, 7]. In addition, it has been adopted as a standard by DBMS suppliers and used in commercial products [16, 14] as well as in a number of research prototypes [2, 25] Technological advances coupled with reduced pricing in hardware have created a new reality in the database field [26] More ....

....servers in such architecture may become the system s bottleneck point as soon as the number of clients attached per cluster increases significantly. Alternative CS DBMS approaches advocate the delegation of database tasks to clients made possible by the available computing resources at this level[11, 10, 6, 8]. The rationale behind these architectures is the off loading of the processing that the database server has to perform. The goal of this paper is to present a range of such clustered database architectures, discuss briefly their performance characteristics, and outline future research directions ....

D. DeWitt, D. Maier, P. Futtersack, and F. Velez. A Study of Three Alternative Workstation--Server Architectures for Object-- Oriented Database Systems. In Proceedings of the 16th Very Large Data Bases Conference, pages 107--121, 1990.

....home page and (2) the relocation time determines the occasion on which a (copied) object is transferred back into its home page. Along both dimensions, we devised an eager and a lazy strategy. To assess the different control strategies, we built an experimental system consisting of a page server [DFMV90] connected with clients managing a segmented dual buffer pool. Our extensive experimental results indicate that a lazy object copying combined with an eager relocation strategy is almost always superior and substantially outperforms pure page based buffer management in most applications. The rest ....

....a duplicate. Shaded granules in Fig. 1 indicate a modification that needs to be flushed upon displacement. Flushing a (shaded) object whose home page is not currently memory resident requires to bring in this page, copy the object into the page and then flush the page. In a client server system [DFMV90] dual buffering can be incorporated in several ways. Kim et al. KBC 88] investigated an object server architecture in which the server buffers pages and ships objects into the client s object oriented workspace. Dual buffering could also be embedded within an object server; i.e. the ....

D. J. DeWitt, P. Futtersack, D. Maier, and F. Velez. A study of three alternative workstation server architectures for object-oriented database systems. In D. McLeod, R. Sacks-Davis, and H. Schek, editors, 16th International Conference on Very Large Data Bases, pages 107--121, Aug 1990.

....and performance aspects in database management systems for engineering environments was already described by several authors from the software engineering as well as the database community. The work presented in this paper differs in several respects from previously published work. DeWitt et al. [9] study object server, page server, and file server architectures for object oriented database systems. They perform their measurements using a benchmark especially developed for this task. This benchmark runs on three prototype systems based on the WiSS [7] which were implemented as test beds for ....

....were implemented as test beds for the measurements. In this paper, we study the communication profile of existing applications using the DBMS GRAS, which is the normal DBMS for these applications. In this way, we hope to measure the exact access profiles our tools impose on the database. Like [9], we focus only on the client server architectures of the systems, while keeping the rest of the DBMS constant. This is in contrast to the work of Chu [8] who considered four different commercial DBMSs to evaluate their use for electronic CAD applications, including relational and ....

[Article contains additional citation context not shown here]

D. DeWitt, P. Futtersack, D. Maier, and F. Velez. A Study of three Alternative Workstation-Server Architectures for Object-Oriented Database Systems. In Proceedings of the Sixteenth Very Large Data Bases Conference, pages 107--121, Brisbane, Australia, 1990.

.... fairly complex and would incur a high overhead, which in particular is not justified when taking typical access patterns in software engineering environments into account (see section 5) Distribution is realized by a variant of the client server approach used by most distributed database systems [20, 22]. However, instead of using one centralized database server, GRAS uses a pool of graph servers. Each of these graph servers controls and manipulates one or more graphs. Whenever an application requests access to a graph, it is connected to a graph server which performs all of its requests for this ....

D.J. DeWitt, D. Maier. A Study of Three Alternative Workstation-Server Architectures for Object-Oriented Database Systems. Proc. 16th Int. Conf. on Very Large Data Bases , IEEE Computer Society Press, pp. 107-121 (1990).

.... [Carey et al. 1993] Dual buffering is also used in [Cheng and Hurson, 1991a] DASDBS [Schek et al. 1990] ORION, O 2 [Bancilhon et al. 1992] and Shore [Carey et al. 1994a] There has been a long discussion in the database literature about the performance advantages of these architectures [DeWitt et al. 1990; Hohenstein Client Server Disk Application Page buffer Object buffer Page buffer Network Objects a a b c d e Figure 2.2: Dual buffer architecture. et al 1997] In cases where network overhead is high and objects cannot be clustered on pages, the object server approach could conceivably be ....

....et al. 1996] made an interesting study in the Thor database in which he compared the performance of a page server system (without prefetching) with an object server system that prefetches groups of objects. The motivation was that the performance of a single object fetching system is unacceptable [DeWitt et al. 1990; Hosking and Moss, 1993] Thor trans 1 Some OODBMSs products, e.g. GemStone, avoid the problem by executing the request on the server and only return the result to the client. fers groups of objects from server to client. On receiving a fetch request, a Thor server selects objects to send ....

[Article contains additional citation context not shown here]

DeWitt, D., Maier, D., Futtersack, P., and Velez, F. (1990). A Study of Three Alternative Workstation-Server Architecture for Object-Oriented Database Systems. In [VLDB, 1990], pages 107--121.

....for clientserver object database management systems (ODBMSs) which can dynamically adapt and operate either as a page server or as an object server. Our fundamental thesis is that page servers and object servers are generally preferable under a limited set of system and workload configurations [DFMV90, CFZ94, DFB 96] and adaptive techniques that combine the features of both are likely to respond better to a larger class of system and workload configurations. Moreover, object servers are not as popular as page servers, This research is supported in part by the Natural Sciences and ....

....Base Endowment. To copy otherwise, or to republish, requires a fee and or special permission from the Endowment. Proceedings of the 25th VLDB Conference Edinburgh, Scotland, 1999 because previous performance studies have shown that object servers are not scalable with respect to data transfer [DFMV90] and concurrency control cache consistency algorithms [CFZ94] In this paper we present new techniques and report performance results that show that objects servers can compete with page servers. 1.1 Background Object servers and page servers are the two competing data shipping architectures ....

[Article contains additional citation context not shown here]

D. DeWitt, P. Futtersack, D. Maier, and F. Velez. A study of three alternative workstation-serverarchitectures for OODBS. In Proceedings of VLDB Conference, 1990.

....many objects, and thus offer means to cluster objects for storage and retrieval, providing hope of attaining our performance goals. We are not committed to client server interaction at a single level of granularity, though, and there is some evidence that multiple granularities need to be offered [14]. We are also not committed to any specific servers or modes of interactions with them; the architecture is designed to be independent of that issue. Application Program Language Run time System (Optional) Mneme Client Library Remote Servers User Process Address Space Local Buffers Server ....

DEWITT, D. J., FUTTERSACK, P., MAIER, D., AND VELEZ, F. A study of three alternative workstation-server architectures for object oriented database systems. In Proceedings of the Sixteenth International Conference of Very Large Data Bases (Brisbane, Australia, Aug. 1990), Morgan Kaufmann, pp. 107--121.

....basis. Physical contiguity is also advantageous in computing environments where objects are moved between client and server machines [Ozsu91] There is some experimental evidence that it is very important to be able to transfer large chunks of data thereby reducing data movement overhead [DeWi90]. Lastly, regarding updates, small changes should have small impact; e.g. inserting few bytes in the middle of the object should not cause the entire object to be re organized. In this paper we present the design of the large object manager of EOS 2 . EOS is a storage system for experimental ....

DeWitt, D.J., D. Maier, P. Futtersack, and F. Velez, "A Study of Three Alternative Workstation-Server Architectures for Object-Oriented Database Systems, " Proc. 16th Int. Conference on Very Large Data Bases, Brisbane, Australia, August 1990, pp. 107-121.

....offloading shared server machines. Data shipping systems can be categorized as page servers, in which clients and servers interact using physical units of data such as pages or segments, and object servers, which send logical units of data such as objects or tuples between clients and servers [DeWitt, 1990] . Many recent systems (e.g. ObServer [Hornick, 1987] O2 [Deux, 1990] EXODUS [EXODUS Project, 1991, Franklin, 1992b] and ObjectStore [Lamb, 1991] have adopted the page server approach due its relative simplicity and potential performance advantages compared to an object server approach. ....

D. DeWitt et al. (1990) A Study of Three Alternative Workstation-Server Architectures for Object-Oriented Database Systems, In Proc. 16th VLDB Conf., Brisbane, Australia.

.... 92] Although there are several alternatives for the granularity of the data items exchanged between the server and the clients, virtually all client server systems have adopted the page server model because of its simplicity and the potential performance advantages over the other alternatives [DMFV90] Under the page server model, the server and the clients interact with each other by exchanging database pages. In a data shipping client server system each client has a buffer pool, also referred to as client cache, where it places the pages fetched from the server. The clients perform most of ....

D. J. DeWitt, D. Maier, P. Futtersack, and F. Velez. A study of three alternative workstationserver architectures for object-oriented database systems. In Proceedings of the Sixteenth International Conference on Very Large Databases, Brisbane, pages 107--121, August 1990.

....With a page server architecture, the objects residing on the same page are also transferred. Different client server architectures can be defined based on how database functionalities are partitioned between the client and the server. Three client server architectures for OODBMSs are described in [6]. For information about the Penguin project, please write to Arthur M. Keller, Stanford University, Computer Science Dept. Stanford, CA 94305 2140, or to ark cs.stanford.edu In contrast to OODBMSs, the interaction between the client and the server in a relational DBMS is based on SQL queries. ....

....from the database. Cache currency is concerned with the effect of database updates on locally cached data, that is, whether an update at the server needs to be propagated to a client cache to ensure cache consistency. The cache consistency issue in client server DBMS architectures is examined in [6, 29, 5, 23]. In [17] we propose predicate based descriptions of client caches to handle the two issues mentioned above for cached objects not identified by logical or physical identifiers. Predicates are obtained from WHERE clauses of SELECT FROM WHERE statements. This paper is organized as follows. In ....

D. J. DeWitt, D. Maier, P. Futtersack, and F. Velez. A Study of Three Alternative Workstation-Server Architectures for ObjectOriented Database Systems. 16th Int. Conf. on Very Large Data Bases, Brisbane, Australia, 1990.

....in turn. 5. 1 Simulation In order to obtain a fair comparison of the the algorithms and an accurate estimate of the cost of garbage collection in a persistent object store, we built a detailed simulator using C CSIM [17] The simulator is based on the client server model of an object server [8] with page level locking. The components of the simulator correspond to those in the persistent object store model of Figure 1. The client buffer pool consists of 150 4KByte pages, and the server pool is 500 4KByte pages. Both the client and server buffer managers use the LRU replacement policy. ....

DeWitt, D.J., Futtersack, P., Maier, D. and Velez, F. "A Study of Three Alternative Workstation-Server Architectures for Object Oriented Database Systems," Proc. 16th International Conference on VLDB, Brisbane, Aug. 1990

....The unit of transfer between the server and the client is a page, i.e. O2 implements a page server architecture [Deux et al. 91] as well as ObjectStore and ONTOS. The server does not need to be able to run user methods for this architecture. O2 used an object server model for a previous version [DeWitt et al. 90] but the overhead was too expensive. 3 User interfaces and development tools ObjectStore offers a choice of a C library interface, a C library interface, or a DML preprocessor interface. The C library interface and C library interface (implemented by the cpp macros facility) preserves the ....

D. DeWitt, P. Futtersack, D. Maier, F. Velez. "A Study of Three Alternative Workstation - Server Architectures for Object Oriented Database Systems", Proceedings of the 16th VLDB Conf., 1990, 107121.

....query is answerable from cache, no communication with the servers is necessary. If a query is only partially answerable from cache, the amount of data needed from the servers may be substantially reduced. 1 SQC provides an alternative approach to page caching and tuple caching architectures [3, 6], in which the unit of transfer between servers and clients is a page or a tuple, respectively. The key advantage of SQC over other approaches is its flexibility: it can be used to answer new queries which are semantically related to previous queries that were cached. 2 Tuple based and ....

D. DeWitt, P. Futtersack, D. Maier, and F. Velez. A study of three alternative workstationserver architectures for object-oriented database systems. In Proceedings of VLDB, 1990.

....contiguous while clustering in memory for network transfer does not. In a page cache, every object is in exactly one cluster and the clustering is the same both on disk and in every memory that caches the object. DeWitt et al. performed the first detailed study of caching in client server systems [35]. They compared a page server that transfers whole pages between client and server and an object server that transfers objects one at a time. Not surprisingly, they found that the page server performed substantially better as long as the clustering of objects on pages was moderately effective. The ....

David J. DeWitt, David Maier, Philippe Futtersack, and Fernando Velez. A study of three alternative workstation-server architectures for object oriented database systems. In Proceedings of the 16th VLDB Conference, pages 107--121, 1990.

....applications, much like 007 which followed it. As with 007, the complex conjunctive and disjunctive queries that we wish to test are not to be found in 001. The suite of queries in 001 doesn t include range search operations either. The HyperModel benchmark [ABM 90] and the ACOB benchmark [DMFV90] also have a traversal focus, as that appears to be a common activity in CAD applications. 6.4.3 Relational Benchmarks The Set Query Benchmark [O N93] was designed to reflect the requirements of Strategic Data Access systems, where queries into large collections of data are used to gain insight. ....

David J. DeWitt, David Maier, Philippe Futtersack, and Fernando Velez. A Study of Three Alternative Workstation-Server Architectures for Object Oriented Database Systems. In Proceedings of the 16th International Conference on Very Large Data Bases, Brisbase, Australia, 1990.

....servers than clients, we expect there to be a bottleneck at the client processor, caused by message handling and query processing overhead. The object server approach has similar difficulties, unless the ability to apply largegrained set operations is available. An example is selection, as in [DFMV90] If the servers can interact in order to resolve inter object references without low granularity client direction that is also an improvement. Without these server features the client must remain a query bottleneck. 5.6 Parallelism In designing a system it is difficult to choose among the above ....

David J. DeWitt, Philippe Futtersack, David Maier, and Fernando Velez. A Study of Three Alternative Workstation-Server Architectures for Object Ori- 158 ented Database Systems. In Proceedings of the 16th VLDB Conference, Brisbane, Australia, 1990.

....be fetched One possibility is to fetch nothing except the needed object; this is called a single object server. Another possibility is to fetch a page containing the needed object and whatever else has been placed on the same page; this is called a page server. An earlier study by De Witt et al. [27] showed that a page server usually outperforms a single object server. There is a third possibility, not considered by that study: it is possible to send a group of objects to the client. Since the size of an average object is on the order of 100 bytes, and a typical hardware page size is 4K bytes ....

....techniques, as implemented in Thor. Context specific or typespecific swizzling appears unjustified. Chapter 5 Prefetching In the previous chapter, the client fetched only one object at a time from a server. This arrangement is simple, but the performance of single object fetching is unacceptable [27, 33, 34]. To improve performance, the number of fetches must be reduced; some entity larger than a single object must be transferred from server to client. This chapter investigates techniques for the server to send more than one object at a time to the client. Although a more accurate name for these ....

[Article contains additional citation context not shown here]

David J. DeWitt, Philippe Futtersack, David Maier, and Fernando Velez. A study of three alternative workstation-server architectures for object oriented database systems. In Proceedings of the 16th Conference on Very Large Data Bases, pages 107--121, Brisbane, Australia, 1990.

....storage system, but also because of the kinds of applications that Object Oriented Database Management Systems (OODBMSs) strive to support, i.e. CAx, GIS, OIS, etc. Furthermore, since QuickStore is implemented on top of the EXODUS Storage Manager (ESM) it is a client server, page shipping system [DeWitt90]. This raises additional performance concerns for recovery that are not present in database systems based on more traditional designs, i.e. centralized DBMSs or systems based on a query shipping architecture. The paper examines four basic recovery techniques that are termed page differencing, ....

....It should be noted that although the recovery algorithms are discussed in the context of QuickStore, they are not QuickStore specific and, in general, are applicable to any similar client server OODBMS. 3.1. Recovery in ESM The EXODUS Storage Manager is a client server, page shipping system [DeWitt90] in which both clients and servers manage their own local buffer pools. When a client needs to access an object on a page that is not currently cached in its local buffer pool, it sends a request (usually over a network) to the appropriate server asking for the page. If necessary, the server reads ....

D. DeWitt, P. Futtersack, D. Maier, F. Velez, "A Study of Three Alternative Workstation-Server Architectures for Object-Oriented Database Systems", Proc. 16th VLDB Conf., Brisbane, Australia, August, 1990.

....the performance of read only transactions, and we expect that detecting updates is much cheaper in QuickStore than in the scheme studied in [Hoski93b] 3. QuickStore Design Concepts As mentioned in Section 1, QuickStore uses ESM to store persistent objects on disk. ESM features a page shipping [DeWitt90] architecture, in which objects are transferred from the server to the client a page at a time. Once a page of objects has been read into the buffer pool of the ESM client, applications that use QuickStore access objects on the page directly in the ESM client buffer pool, by dereferencing normal ....

D. DeWitt, P. Futtersack, D. Maier, F. Velez, "A Study of Three Alternative Workstation-Server Architectures for Object-Oriented Database Systems", Proceedings of the 16th International Conferece on Very Large Data Bases, Brisbane, Australia, August, 1990.

....for semantic object clustering (discussed further below) its very simple schema, and its lack of complex operations such as traversals. Also in this area, researchers at Altair designed a complex object benchmark (ACOB) for use in studying alternative client server process architectures [DFMV90] Unlike previous OODB benchmarks, a notion of complex objects was included in the design; however, only a small number of traversal and update operations were involved, as these were sufficient to expose the tradeoffs among the software architectures of interest. Finally, Winslett and Chu ....

....E Exodus Exodus consists of two main components: The Exodus Storage Manager (ESM) and the E programming language. The ESM provides files of untyped objects of arbitrary size, B trees, and linear hashing. The current version of the ESM (Version 2. 2) EXO92] uses a page server architecture [DFMV90] where client processes request pages that they need from the server via TCP IP. If the server cannot satisfy the request from its buffer pool, a disk I O is initiated by invoking a disk process to perform the actual I O operation. After the disk process has read in the page, the 11 server ....

[Article contains additional citation context not shown here]

David J. DeWitt, Philippe Futtersack, David Maier, and Fernando Velez. A study of three alternative workstation-server architectures for object-oreinted database systems. In Proceedings of the VLDB Conference, Brisbane, Australia, August 1990.

....on complex object assembly and not on general query processing. Our solution was to develop a smaller group of benchmarks that focus on clustering, buffer size, window size, database size and scheduling algorithms. Our benchmark most closely resembles the Altair Complex Object Benchmark (ACOB) [22]. Each complex object is structured as a binary tree of 3 levels. However, unlike objects in ACOB, our objects are physically stored as a single record, not a group of seven records. Each object consists of 4 integer and 8 object reference fields equaling 96 bytes, resulting in 9 objects per page. ....

D.J. DeWitt, P. Futtersack, D. Maier, and F. Velez, "A Study of Three Alternative Workstation-Server Architectures for Object-Oriented Database Systems," Sixteenth International Conference on Very Large Data Bases, p. 107 (1990).

No context found.

David J. DeWitt, David Maier, Philippe Futtersack, and Fernando Velez. A Study of Three Alternative Workstation-Server Architectures for Object-Oriented Database Systems. In Proceedings of the 16th International Conference on Very Large Databases, Brisbane, Australia (VLDB '90), pages 107--121, August 1990.

No context found.

D. DeWitt, D. Maier, P. Futtersack, and F. Velez. A Study of Three Alternative Workstation--Server Architectures for Object--Oriented Database Systems. In Proceedings of the 16th International Conference on Very Large Data Bases, pages 107--121, Brisbane, Australia, 1990.

No context found.

David J. DeWitt, Philippe Futtersack, David Maier, and Fernando Velez. "A Study of Three Alternative Workstation-Server Architectures for Object Oriented Database Systems", Proceedings of the 16th VLDB Conference, Brisbane, Australia, 1990, pp. 107-121.

No context found.

DeWitt, D., P. Futtersack, D. Maier, and F. Velez: 1990, `A study of three alternative workstation-server architectures for OODBS'. In: Proc. 16th International Conference on Very Large Data Bases. pp. 107--121.

No context found.

D. J. DeWitt, D. Maier, P. Futtersack, F. Velez. A study of three alternative workstation-server architectures for object-oriented database systems. VLDB 1990, pages 107-121.

No context found.

D. J. DeWitt, D. Maier, P.Futtersack, and F. Velez. A study of three alternative workstation-server architectures for object-oriented database systems. In Proceedings of the Sixteenth International ConferenceonVery Large Databases, Brisbane, pages 107#121, August 1990.

No context found.

D. J. DeWitt, D. Maier, P.Futtersack, and F. Velez. A Study of Three AlternativeWorkstation -Server Architectures for Object-Oriented Database Systems. In Proceedings of the Sixteenth International ConferenceonVery Large Databases, Brisbane, pages 107#121, August 1990.

No context found.

DeWitt, D.J., D. Maier, P. Futtersack, and F. Velez, "A Study of Three Alternative Workstation-Server Architectures for Object-Oriented Database Systems," Proc. 16th Int. Conference on Very Large Data Bases, Brisbane, Australia, August 1990, pp. 107121.

First 50 documents  Next 50

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