José Orlando Pereira

There is nowadays an increasing need for database replication, as the construction of high performance, highly available, and large-scale applications depends on it to maintain data synchronized across multiple servers. A particularly popular approach, used for instance byFacebook, is the MySQL open source database management system and its built-in asynchronous replication mechanism. The limitations imposed by MySQL on replication topologies mean that data has to go through a number of hops or each server has to handle a large number of slaves. This is particularly worrisome when updates are accepted by multiple replicas and in large systems. It is however difficult to accurately evaluate the impact of replication in data freshness, since one has to compare observations at multiple servers while running a realistic workload and without disturbing the system under test. In this paper we address this problem by introducing a tool that can accurately measure replication delays for any workload and then apply it to the industry standard TPC-C benchmark. This allows us to draw interesting conclusions about the scalability properties of MySQL replication.

The Database Management System (DBMS) used to be a commodity software component, with well known standard interfaces and semantics. However, the performance and reliability expectations being placed on DBMSs have increased the demand for a variety add-ons, that augment the functionality of the database in a wide range of deployment scenarios, offering support for features such as clustering, replication, and selfmanagement, among others. The effectiveness of such extensions largely rests on closely matching the actual needs of applications, hence on a wide range of tradeoffs and configuration options out of the scope of traditional client interfaces. A well known software engineering approach to systems with such requirements is reflection. Unfortunately, standard reflective interfaces in DBMSs are very limited (for instance, they often do not support the desired range of atomicity guarantees in a distributed setting). Some of these limitations may be circumvented by implementing reflective features as a wrapper to the DBMS server. Unfortunately, this solutions comes at the expense of a large development effort and significant performance penalty. In this paper we propose a general purpose DBMS reflection architecture and interface, that supports multiple extensions while, at the same time, admitting efficient implementations. We illustrate the usefulness of our proposal with concrete examples, and evaluate its cost and performance under different implementation strategies.