THIS SITE

Information for

2017 technical reports

CS-2017-01
Title

Revisiting the CLBlood Model: Formulation Enhancements and Online Deployment

Authors

Spencer R. Van Leeuwen, Gladimir V. G. Baranoski and Bradley W. Kimmel

Abstract

CLBlood is a cell-based light interaction model for human blood previously developed by the Natural Phenomena Simulation Group (NPSG) at the University of Waterloo. In this report, we revisit several elements of its formulation and describe appropriate enhancements. Furthermore, we compare our current and previous results to demonstrate that the predictive capabilities of our model have been fully preserved in its revised version. We also showcase the online deployment of CLBlood.

Date January 27, 2017
Report CS-2017-01 (PDF)
CS-2017-02
Title

Renormalization of NoSQL Database Schemas

Authors

Michael J. Mior, Kenneth Salem

Abstract

Applications using a NoSQL database may duplicate attribute values in multiple physical structures in order to provide additional access paths. This is necessary in order to meet performance goals but introduces complications. Typically, the NoSQL DBMS has no understanding of application-level denormalization. This means that if a user wishes to issue ad-hoc queries, it is necessary to understand this denormalization and to manually select which physical structures to use for a given query. Furthermore, if application requirements change, this often forces a change in the schema. This can impact existing application code which is no longer valid under the previous physical design.

We propose an algorithm for reconstructing a normalized logical schema from a denormalized physical design. Our method begins with a description of a denormalized physical schema along with functional and inclusion dependencies on this schema. By combining inference based on these dependencies with traditional normalization techniques, we produce a schema normalized according these dependencies which applications can be use as a logical model. This logical schema facilitates transparent optimization of ad-hoc queries using existing physical structures. We enable this optimization by also producing a mapping between the physical and logical schema as part of our normalization process. Furthermore, this logical schema is also useful as input to physical design tools in order to optimize the physical schema.

Date April 3, 2017
Report CS-2017-02 (PDF)
CS-2017-03
Title

Database Managed CPU Performance Scaling for Improved Energy Efficiency

Authors

Mustafa Korkmaz, Martin Karsten, Kenneth Salem, Semih Salihoglu

Abstract

Dynamic voltage and frequency scaling (DVFS) is a technique for adjusting the speed and power consumption of processors, allowing performance to be traded for reduced power consumption. Since CPUs are typically the largest consumers of power in modern  servers, DVFS can have a significant impact on overall server power consumption. Modern operating systems include DVFS governors, which interact with the processor to manage performance and power consumption according to some system-level policy.

In this paper, we argue that for database servers, DVFS can be managed more effectively by the database management system. We present a power-aware database request scheduling algorithm called POLARIS. Unlike operating system governors, POLARIS is aware of database units of work and database performance targets, and can achieve a better power/performance tradeoff by exploiting this knowledge. We implemented POLARIS in SHORE-MT, and we show that it can improve both power consumption and performance relative to operating system baselines.

Date June 2017
Report CS-2017-03 (PDF)