Waqar Hasan

Query optimization for parallel execution

The decreasing cost of computing makes it economically viable to reduce the response time of decision support queries by using parallel execution to exploit inexpensive resources. This goal poses the following query optimization problem: Minimize response time subject to constraints on throughput, which we motivate as the dual of the traditional DBMS problem. We address this novel problem in the context of Select-Project-Join queries by extending the execution space, cost model and search algorithm that are widely used in commercial DBMSs. We incorporate the sources and deterrents of parallelism in the traditional execution space. We show that a cost model can predict response time while accounting for the new aspects due to parallelism. We observe that the response time optimization metric violates a fundamental assumption in the dynamic programming algorithm that is the linchpin in the optimizers of most commercial DBMSs. We extend dynamic programming and show how optimization metrics which correctly predict response time may be designed.

The Iris architecture and implementation

The goals of the Iris database management system are to enhance database programmer productivity and to provide generalized database support for the integration of future applications. Iris is based on an object and function model. Iris objects are typed but unlike other object systems, they contain no state. Attribute values, relationships and behavior of objects are modeled by functions. The Iris architecture efficiently supports the evaluation of functional expressions. The goal of the architecture is to provide a database system that is powerful enough to support the definition of functions and procedures that implement the semantics of the data model. An overview of the data model is provided, the architecture is described in detail, and implementation experience and usage of the system are discussed. >

Scheduling problems in parallel query optimization

We introduce a class of novel multiprocessor scheduling problems that arise in the optimization of SQL queries for parallel machines. These consist of scheduling a tree of interdependent communicating operators while exploiting both inter-operator and intra-operator parallelism. We develop algorithms for the specific problem of scheduling a Pipelined Operator Tree in which all operators run in parallel using inter-operator parallelism. Weights associated with nodes and edges represent respectively the cost of operators and communication. Communication cost is incurred only if adjacent operators are assigned different processors. The optimization problem is to assign operators to processors so as to minimize the maximum processor load. We develop two approximation algorithms for this NP-hard problem. The faster algorithm has a performance ratio of 3.56 while the slower algorithm has a ratio of 2.87.

Open issues in parallel query optimization

We provide an overview of query processing in parallel database systems and discuss several open issues in the optimization of queries for parallel machines.

Parallelism and its price: a case study of nonstop SQL/MP

We describe the use of parallel execution techniques and measure the price of parallel execution in NonStop SQL/MP, a commercial parallel database system from Tandem Computers. Non-Stop SQL uses intra-operator parallelism to parallelize joins, groupings and scans. Parallel execution consists of starting up several processes and communicating data between them. Our measurements show (a) Startup costs are negligible when processes are reused rather than created afresh (b) Communication costs are significant — they may exceed the costs of operators such as scan, grouping or join. We also show two counter-examples to the common intuition that parallel execution reduces response time at the expense of increased work — parallel execution may reduce work or may increase response time depending on communication costs.

The papyrus integrated data server

Summary form only given. The authors focus on the performance of integrated specialized data managers. In particular, they focus on the customizations of parallel executions of data manager operators in a variety of computer configurations. This is done by specifying the glue that connects data manager operators in a way that is independent of the computer configuration; and then providing the ability to transparently target the execution to a variety of computer configurations. Parallelization of data manager operators built using Papyrus services is a challenging problem, but a more challenging problem is the parallelization of data manager operators that are built independently of Papyrus and are therefore black boxes to Papyrus. Papyrus is a set of modules and services that enables the parallelization and integration of specialized data managers into one execution environment. Papyrus programs can be transparently targeted to different hardware configurations and can dynamically adjust at runtime to the number of available resources. A Papyrus System consists of a number of clients interfacing to a Papyrus Server. The Server consists of several integrated data managers executing on a multiprocessor system.<<ETX>>

PIL: An Optimizable Functional Language for Data Intensive Applications

The Papyrus Interface Language (PIL) has the design goal of providing optimizable and parallelizable language features. Analogous to the design philosophy of RISC instruction sets, the design of PIL is motivated by the desire to exploit query optimization and parallelization techniques. In contrast, most proposals of database programming language provide features to directly match user needs irrespective of the implementation problems, analogous to the CISC instruction set proposals. We have combined a functional model of computation with data types suitable for data intensive applications. A functional model gives a declarative semantics to all expressions including “procedural” constructs such as if-then-else, while and function calls provided the expression is without side-effects. We also provide specialized constructs for iteration over bags and sequences in order to facilitate optimization. The semantics of data types and computational abstractions are carefully chosen to retain the capability to parallelize programs. We have chosen to only partially define the order of evaluation for programs. This opens up more opportunities for reordering and for parallel execution. Just as in RISC instruction sets — what is not included will be a factor in dictating the performance of the system — we argue the need to exclude features such as object identity, semantic types an d inheritance that are popularly included in most database programming languages.

The Iris kernel architecture

We describe an architecture for a database system based on an object/function model. The architecture efficiently supports the evaluation of functional expressions. The goal of the architecture is to provide a database system that is powerful enough to support the definition of functions and procedures that implement the semantics of the data model. The architecture has been implemented to support the Iris Database System.

Architectural Concepts for Large Knowledge Bases

We describe experience in structuring knowledge gained in two parallel projects. Earlier KBMS projects focused on simplifying access and improving the performance of databases using concepts from artificial intelligence. The Rx project developed methods to extract knowledge from databases. Current research (KSYS) exploits this experience to develop and demonstrate new concepts and techniques for managing large collections of knowledge and data.