A Tool for Performance-Driven Design of Parallel Systems

Boughter, Ellen; Alexander, William; Keller, Teresa

doi:10.1007/978-1-4613-0533-0_10

Cited by 6 publications

(2 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The metric for evaluating our data placement strategy is transaction throughput. To estimate throughput, we implemented a modeling tool called FIRM, which is described in detail in [Bou87], FIRM consists of three programs: the first performs the data-placement algorithm, the second maps workloads to individual IRs and the third solves the resulting analytic queueing model. Inputs to FIRM are: 1) the Order_Entry workloads, 2) the Bubba architecture and configuration description, and 3) data-placement parameters (e.g., CDegDecl).…”

Section: The Performance Modelmentioning

confidence: 99%

Data placement in Bubba

et al. 1988

Self Cite

View full text Add to dashboard Cite

This paper examines the problem of data placement in Bubba, a highly-parallel system for data-intensive applications being developed at MCC. “Highly-parallel” implies that load balancing is a critical performance issue. “Data-intensive” means data is so large that operations should be executed where the data resides. As a result, data placement becomes a critical performance issue. In general, determining the optimal placement of data across processing nodes for performance is a difficult problem. We describe our heuristic approach to solving the data placement problem in Bubba. We then present experimental results using a specific workload to provide insight into the problem. Several researchers have argued the benefits of declustering (i e, spreading each base relation over many nodes). We show that as declustering is increased, load balancing continues to improve. However, for transactions involving complex joins, further declustering reduces throughput because of communications, startup and termination overhead. We argue that data placement, especially declustering, in a highly-parallel system must be considered early in the design, so that mechanisms can be included for supporting variable declustering, for minimizing the most significant overheads associated with large-scale declustering, and for gathering the required statistics.

show abstract

Section: The Performance Modelmentioning

confidence: 99%

Data placement in Bubba

et al. 1988

Self Cite

View full text Add to dashboard Cite

show abstract

“…To estimate throughput, we implemented a modeling tool called FIRM, which is described in detail in [Bou87], FIRM consists of three programs: the first performs the data-placement algorithm, the second maps workloads to individual IRs and the third solves the resulting analytic queueing model. Inputs to FIRM are: 1) the Order_Entry workloads, 2) the Bubba architecture and configuration description, and 3) data-placement parameters (e.g., CDegDecl).…”

Section: The Performance Modelmentioning

confidence: 99%

Data placement in Bubba

Copeland

Alexander

Boughter

et al. 1988

Proceedings of the 1988 ACM SIGMOD International Conference on Management of Data - SIGMOD '88

Self Cite

150

View full text Add to dashboard Cite

This paper examines the problem of data placement in Bubba, a highly-parallel system for data-intensive applications being developed at MCC. "Highly-parallel" implies that load balancing is a critical performance issue. "Data-intensive" means data is so large that operationsshould be executed where the data resides. As a result, data placement becomes a critical performance issue.In general, determining the optimal placement of data across processing nodes for performance is a difficult problem. We describe our heuristic approach to solving the data placement problem in Bubba. We then present experimental results using a specific workload to provide insight into the problem. Several researchers have argued the benefits of declustering (i.e., spreading each base relation over many nodes). We show that as declustering is increased, load balancing continues to improve. However, for transactions involving complex joins, further declustering reduces throughput because of communications, startup and termination overhead.We argue that data placement, especially declustering, in a highly-parallel system must be considered early in the design, so that mechanisms can be included for supporting variable declustering, for minimizing the most significant overheads associated with large-scale declustering, and for gathering the required statistics.

show abstract