D. Schneider, "Complex query processing in multiprocessor database machines, PhD thesis," Computer Sciences Technical Report 965, Universiteit of Wisconsin, Madison, USA, September 1990.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....5: Bushy tree with its right deep segments. 3.3 Segmented Right Deep execution (liD) In contrary to SE, segmented right deep execution uses interoperator pipelining in addition to intra operator parallelism. This strategy is proposed in [CLY92] a paper which was inspired by [ScD90] Schneider [Sch90,ScD90] describes the differences in possible parallelism between left deep and right deep linear join trees , when the simple hash join is used for the individual join operations. In a right deep tree the buildphases of all join operations can be executed in parallel and after that probe phases can be ....

....integer attributes, and after each join they are projected to the second integer attributes and the remaining attributes of one of the operands, so that the result of each operation again is a Wisconsin relation equal in size to the operands. This test problem is similar to the problem used in [Sch90], in [ZZS93] and in [WiA93] All possible join trees for this query have the same total execution costs. Also, the individual join operations are equal in costs and sizes of its operands. So, any differences in response time are caused by differences in the shape of the tree and the ....

[Article contains additional citation context not shown here]

D. Schneider, "Complex query processing in multiprocessor database machines, PhD thesis," Computer Sciences Technical Report 965, Universiteit of Wisconsin, Madison, USA, September 1990.

....of operators on adjacent nodes in producerconsumer mode. Pipelining is determined by the structure of the query processing tree [16] In the left deep tree used for instance in System R [23] the output of each node is materialized before being read by the parent node. In the right deep tree [22], the right stream input to a node is consumed in pipeline mode, while the left input stream must still be materialized. Several variations of the right deep tree have been proposed [l] 36] to alleviate its main disadvantage, namely the high memory demand. For the full exploitation of bushy ....

D.A. Schneider, "Complex Query Processing in Multiprocessor Database Machines," Technical Report TR965, Univ. of Wisconsin, Madison, 1990.

....their corresponding costs into consideration was proposed. A two step approach to deal with join sequence scheduling and processor allocation for parallel query processing was devised in [6] Several query plans in processing multi join queries in a shared nothing architecture were investigated in [27]. In addition, experimental studies on evaluating various query plan generation strategies were conducted in [36] Among various join methods, the hash join has been the focus of much research effort and reported to have performance superior to that of others, particularly because it presents an ....

D. Schneider. Complex Query Processing in Multiprocessor Database Machines. Technical Report Tech. Rep. 965, Computer Science Department, University of Wisconsin-Madison, September 1990.

....the resulting relation from joining the two relations with its two child vertices, and the query tree is executed bottom up. Conventionally, in the context of hash joins, the left and right child vertices of an internal vertex denote, respectively, the inner and outer relations of a join [33], where, as explained in Section 2, the inner relation is the relation used to build the hash table and the outer relation is the one whose tuples are applied to probe the hash table. Examples of the three forms of query trees are shown in Figure 1, where the inner and outer relations are ....

....forms of query trees are shown in Figure 1, where the inner and outer relations are indicated for illustration. It can be seen that both right deep and bushy trees allow the implementation of pipelining. Schneider and DeWitt are among the first to study the effect of pipelining for hash joins [33] [35] where the focus was on the use of right deep trees due mainly to the simplicity of right deep trees and the uncertainty for the improvement achievable by using bushy trees. Clearly, for a given query, the number of right deep trees to be considered is significantly less than that of bushy ....

[Article contains additional citation context not shown here]

D. Schneider. Complex Query Processing in Multiprocessor Database Machines. Technical Report Tech. Rep. 965, Computer Science Department, University of Wisconsin-Madison, September 1990.

....using sort merge joins was devised in [7] Pipelining hash joins in a bushy tree and processor allocation within each pipeline were studied in [5] and [18] respectively. In addition, various query plans in processing multi join queries in a shared nothing architecture were investigated in [24]. While most prior work on inter operator parallelism focused on the execution tree generation to minimize the query execution cost, there is relatively little result reported on exploiting the structure of a query tree to further reduce each individual join cost. It has been shown that the cost ....

D. Schneider. Complex Query Processing in Multiprocessor Database Machines. Technical Report Tech. Rep. 965, Computer Science Department, University of Wisconsin-Madison, September 1990.

....scheduling and processor allocation for parallel query processing was devised in [5] A hierachical approach was proposed in [27] to schedule the execution of multiple queries. In addition, various query plans in processing multi join queries in a shared nothing architecture were studied in [19] [21] Among various join methods, the hash join has been the focus of much research effort and reported to have performance superior to that of others, particularly because it presents an opportunity for pipelining [6] 17] 18] 26] A pipeline of hash joins is composed of several stages, each ....

D. Schneider. Complex Query Processing in Multiprocessor Database Machines. Technical Report Tech. Rep. 965, Computer Science Department, University of Wisconsin-Madison, September 1990.

....node represents the resulting relation from joining the two relations with its two child nodes, and the query tree is executed in a manner of bottom up. In the context of hash joins the left and right child nodes of an internal node denote, respectively, the inner and outer relations of a join [19], where, as explained in Section 2, the inner relation is the relation used to build the hash table and the outer relation is the one whose tuples are applied to probe the hash table. Examples of the three forms of query trees are shown in Figure 1, where the inner and outer relations are ....

....of the three forms of query trees are shown in Figure 1, where the inner and outer relations are indicated for illustration. It can be seen that both right deep and bushy trees allow the implementation of pipelining. Schneider and DeWitt are among the first to study the effect of pipelining [19] [21] where the focus was on the use of right deep trees due mainly to the simplicity of right deep trees and the uncertainty for the improvement achievable by using bushy trees. Clearly, for a given query, the number of right deep trees to be considered is significantly less than that of bushy ....

[Article contains additional citation context not shown here]

D. Schneider. Complex query processing in multiprocessor database machines. Technical Report Tech. Rep. 965, Computer Science Department, Univ. Wisconsin-Madison, Sep. 1990.

....mentioned above, most efforts are experimental in nature and offer no theoretical justification for the algorithms that they propose. In addition, many proposals have simplified the scheduling issues by ignoring independent (bushy tree) parallelism; these include the right deep trees of Schneider [Sch90a] and the segmented right deep trees of Chen et al. CLYY92] Nevertheless, the advantages offered by such parallelism, especially for large queries, have been demonstrated in prior research [CYW92] Tan and Lu [TL93] and Niccum et al. NSHL95] consider the general problem of scheduling bushy join ....

....discussion to s = 1 (i.e. memory) An obvious advantage of this general formulation is that it allows us the flexibility to draw the line between ts and ss resources at any boundary, depending on factors such as application requirements or user view of resources. An operator tree [GHK92, Hon92, Sch90a] is created as a macro expansion of an execution plan tree by refining each node into a subtree of physical operator nodes, e.g. scan, probe, build (Figure 2(a,b) Edges represent the flow of data as well as two forms of timing constraints between operators: pipelining (thin edges) and ....

[Article contains additional citation context not shown here]

Donovan A. Schneider. "Complex Query Processing in Multiprocessor Database Machines". PhD thesis, University of Wisconsin-Madison, September 1990.

....using sort merge joins was devised in [6] Pipelining hash joins in a bushy tree and processor allocation within each pipeline were studied in [4] and [15] respectively. In addition, various query plans in processing multi join queries in a sharednothing architecture were investigated in [20]. While most prior work on inter operator parallelism focused on the execution tree generation to minimize the query execution cost, there is relatively little result reported on exploiting the structure of a query tree to further reduce each individual join cost. It has been shown that the cost ....

D. Schneider. Complex Query Processing in Multiprocessor Database Machines. Technical Report Tech. Rep. 965, Computer Science Department, University of Wisconsin-Madison, September 1990.

....using sort merge joins was devised in [7] Pipelining hash joins in a bushy tree and processor allocation within each pipeline were studied in [5] and [18] respectively. In addition, various query plans in processing multi join queries in a shared nothing architecture were investigated in [24]. While most prior work on inter operator parallelism focused on the execution tree generation to minimize the query execution cost, there is relatively little result reported on exploiting the structure of a query tree to further reduce each individual join cost. It has been shown that the cost ....

Schneider D (1990) Complex query processing in multiprocessor database machines. Tech Rep 965, Computer Science Department, University of Wisconsin, Madison

....of the 20th VLDB Conference Santiago, Chile, 1994 PMC 90] Two communicating operators may either incur the communication overhead and run on distinct processors, or share a processor and save the communication overhead. We model parallelization as scheduling an operator tree [GHK92, Hon92b, Sch90] on a parallel machine. We represent the resource requirements of operators and communication as node and edge weights respectively. Our optimization objective is to find a schedule (i.e. a parallel plan) with minimal response time. This paper concentrates on algorithms for exploiting pipelined ....

....resources to operators. We consider parallelization itself to consist of two steps. The first step translates an annotated join tree to an operator tree. The second step is a scheduling step that allocates resources to operators. Section 2. 1 refines prior notions of operator trees [GHK92, Hon92b, Sch90] by reducing timing constraints between operators to parallel and precedence constraints. Section 2.2 shows how resource requirements of nodes and edges may be derived from conventional cost models. Section 2.3 describes a cost model for response time. Finally, Section 2.4 provides a formal ....

D. A. Schneider. Complex Query Processing in Multiprocessor Database Machines. PhD thesis, University of Wisconsin---Madison, September 1990. Computer Sciences Technical Report 965.

.... as a collection of modules in the DeNet simulation language [Livn89] The simulator was derived from a single site configuration of a simulator constructed for the Gamma parallel database system [DeWi90] and used in studies of replication strategies [Hsai90] and complex query processing [Schn90]. We used this simulator as a starting point primarily to facilitate subsequent research on MV2PL extended for use in a parallel DBMS environment. In addition, the basic Gamma simulator was validated against the actual Gamma implementation [Schn90, Hsai90] so we knew that we were at least ....

....strategies [Hsai90] and complex query processing [Schn90] We used this simulator as a starting point primarily to facilitate subsequent research on MV2PL extended for use in a parallel DBMS environment. In addition, the basic Gamma simulator was validated against the actual Gamma implementation [Schn90, Hsai90], so we knew that we were at least starting from something that modeled reality fairly accurately. In order to explain the model, we will break it down into two major components, the application model and the system model. Each of these have several subcomponents that we will describe in this ....

Schneider, D., Complex Query Processing in Multiprocessor Database Machines, Ph.D. Thesis, Computer Sciences Department, University of Wisconsin-Madison, September 1990.

.... One of the most difficult issues in database parallelism is the efficient exploitation of pipelining, because it does not easily lend itself to load balancing , as pointed out by Graefe [Gra93] Different aspects of pipeline exploitation on various query tree structures are studied in [Sch90, GHK92, LVZ93, SYT93, ZZBS93] and others. However, in all those models, the notion of pipeline is rather restricted, since, even for bushy trees, at most one of the children of a node is assumed to produce output in pipeline mode. While this is acceptable for some algorithms implementing relational algebra operators, it is ....

Donovan A. Schneider. Complex query processing in multiprocessor database machines. Technical Report TR965, University of Wisconsin, Madison, Wisconsin, 1990.

....mentioned above, most efforts are experimental in nature and offer no theoretical justification for the algorithms that they propose. In addition, many proposals have simplified the scheduling issues by ignoring independent (bushy tree) parallelism; these include the right deep trees of Schneider [Sch90] and the segmented right deep trees of Chen et al. CLYY92] Nevertheless, the advantages offered by such parallelism, especially for large queries, have been demonstrated in prior research [CYW92] Tan and Lu [TL93] and Niccum et al. NSHL95] consider the general problem of scheduling bushy join ....

....to be time sliceable or preemptable, in the sense that they can be time shared among different operations at low overhead. Resources like the CPU(s) the disk(s) and the network interface(s) or communication processor(s) are preemptable, while memory is not. An operator tree [GHK92, Hon92, Sch90] is created as a macro expansion of an execution plan tree by refining each node into a subtree of physical operator nodes, e.g. scan, probe, build (Figure 1(a,b) Edges represent the flow of data as well as two forms of timing constraints between operators: pipelining (thin edges) and ....

Donovan A. Schneider. "Complex Query Processing in Multiprocessor Database Machines". PhD thesis, University of Wisconsin-Madison, September 1990.

....mentioned above, most efforts are experimental in nature and offer no theoretical justification for the algorithms that they propose. In addition, many proposals have simplified the scheduling issues by ignoring independent (bushy tree) parallelism; these include the right deep trees of Schneider [Sch90] and the segmented right deep trees of Chen et al. CLYY92] Nevertheless, the advantages offered by such parallelism, especially for large queries, have been demonstrated in prior research [CYW92] Tan and Lu [TL93] and Niccum et al. NSHL95] consider the general problem of scheduling bushy join ....

....to be time sliceable or preemptable, in the sense that they can be time shared among different operations at low overhead. Resources like the CPU(s) the disk(s) and the network interface(s) or communication processor(s) are preemptable, while memory is not. An operator tree [GHK92, Hon92, Sch90] is created as a macro expansion of an execution plan tree by refining each node into a subtree of physical operator nodes, e.g. scan, probe, build (Figure 1(a,b) Edges represent the flow of data as well as two forms of timing constraints between operators: pipelining (thin edges) and ....

D. A. Schneider. "Complex Query Processing in Multiprocessor Database Machines". PhD thesis, University of Wisconsin-Madison, September 1990.

....not produce a response time optimal join plan. Thus, the complexity of the query optimization problem is harder in a parallel RDBMS than in a single site RDBMS, which is already an NP Hard problem. Recognizing this complexity, past work on this topic has focused on finding heuristic algorithms [2, 7], or on outlining a general algorithm with high complexity [3] Also, there has been work on scheduling and processor allocation of query tasks assuming a optimal query plan has already been generated [1] Yu et al. [9] review some aspects of parallel query processing but have not studied the ....

....response time is minimized among all schedules which follow the same partial join order. In other words, we build in the scheduling into our optimization phase to get optimal response time solutions. This is in contrast with a twostep approach (optimization followed by scheduling the optimal plan) [1, 7] which unfortunately does not yield to optimal solutions. We make several assumptions that simplify our presentation somewhat. Most of these are either not very restrictive, or relatively easy to relax. We do not consider compatible partitioning which says that two (intermediate) relations are ....

Donovan A. Schneider. Complex Query Processing in Multiprocessor Database Machines. PhD thesis, University of Wisconsin-Madison, 1990.

....execution [DG92, Val93] to speed up database queries presents a parallelism communication trade off. While work is divided among processors, the concomitant communication increases total work itself [Gra88, PMC 90] We can represent the task to be scheduled as a weighted operator tree [Hon92, Sch90, HM94a] in which nodes represent atomic units of execution (operators) and directed edges represent the flow of data as well as timing constraints between operators. E mail: chekuri cs.stanford.edu. Supported by an OTL grant and NSF Young Investigator Award CCR 9357849, with matching funds ....

....that fixes aspects such as the order of joins and the strategy for computing each join. The second phase, parallelization, converts the annotated query tree into a parallel plan. Parallelization itself has two steps. The first converts the annotated query tree to an operator tree [GHK92, Hon92, Sch90] The second schedules the operator tree on a parallel machine. In this paper, we are only concerned with the second phase. Several approaches exist for the first phase; Hong and Stonebraker [HS91] used a conventional query optimizer while Hasan and Motwani [HM95] develop algorithms that ....

D. A. Schneider. Complex Query Processing in Multiprocessor Database Machines. PhD thesis, University of Wisconsin---Madison, September 1990. Computer Sciences Technical Report 965.

....to represent the initial query [12] The simplest type is the left deep tree, used e.g. in System R and R [17, 14] This tree type does not allow pipelining. Also, since at most one child of any node can be another operator, bushy parallelism cannot be exploited either. In the right deep tree [16], all operators are executed in pipeline mode, but the memory demand is high. The zigzag tree [25] alleviates this problem by combining left deep and right deep subtrees. Right deep stratified trees [1] exploit pipelining, while supporting also bushy parallelism to a limited degree. However, bushy ....

D. A. Schneider. Complex query processing in multiprocessor database machines. Technical Report TR965, University of Wisconsin, Madison, Wisconsin, 1990.

....of choosing left deep and bushy trees. A two step approach to deal with join sequence scheduling and processor allocation for parallel query processing was devised in [2] In addition, various query plans in processing multi join queries in a sharednothing architecture were investigated in [10]. Among various join methods, the hash join has been the focus of much research effort and reported to have performance superior to that of others, particularly because it presents an opportunity for pipelining. A pipeline of hash joins is composed of several stages, each of which is associated ....

....has been shown to be very effective in reducing the query execution time, prior studies on pipelined hash joins have focused mainly on heuristic methods for query plan generation. Most of the prior work on query plan generation, such as static right deep scheduling, dynamic bottom up scheduling [10], and segmented right deep trees, resorted to simple heuristics to allocate processors to pipeline stages. Also, due to the shared nothing architecture assumed, prior methods mainly dealt with memory as a constraint for the execution of pipelined hash joins. Little effort was made to take ....

D. Schneider. Complex Query Processing in Multiprocessor Database Machines. Technical Report Tech. Rep. 965, Computer Science Department, UW-Madison, September 1990.

No context found.

D. Schneider. Complex Query Processing in Multiprocessor Database Machines. Technical Report Tech. Rep. 965, Computer Science Department, University of Wisconsin-Madison, September 1990.

No context found.

Schneider, D.A. Complex Query Processing in Multiprocessor Database Machines. Ph.D. thesis, University of Wisconsin, Madison, Sept. 1990. Computer Sciences Technical Report 965.

No context found.

D. Schneider, Complex Query Processing in Multiprocessor Database Machines, PhD thesis, University of Wisconsin, Madison, 1990.

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