Performance tradeoffs for client-server query processing

Abstract

The construction of high-per formancedatabuse systems that combine the best aspects of the relational and object-oriented approaches requires the design of client-server architectures that can fully exploit client and~erver resources in a jlexible manner The two predominant paradigms for c[ient-server query execution are datashipping and query -shqrpirrg We jirs~dejine these pohcies m wrms of the restrictions thev place on operator site se[ection during query optimization. We then investigate the performance tradeoffs between them for bulk query processing.While each strategy has advantages, nei!her one on its own is eflicient across a wide range of circumstances. We describe andevaluate a more~exible policy called hybrid-shipping, which can execute queries at clients, servers, or any combination of the two. Hybrid-shipping is shown to at least match the best of the two "pure" policies, and in some situations, to perform better than both. The implementation of hybrid-shipping raises a number of dlficuh problems for query optimization. We describe an initial investigation into the use of a .2-step query optimization strategy as a way of addressing these issues.

Just-In-Time Data Distribution for Analytical Query Processing

“…Caching has been traditionally used to improve performance in distributed systems [7,13]. Similarly to hybrid-shipping query processing in [7] Octopus allocates at run time subplans to workers based on the current cache content.…”

“…Similarly to hybrid-shipping query processing in [7] Octopus allocates at run time subplans to workers based on the current cache content. However, JIT data shipping is not a processing by-product, but an intentional act performed for the purpose of parallel processing on non-dedicated machines.…”

Don’t Reveal My Intension: Protecting User Privacy Using Declarative Preferences during Distributed Query Processing

“…Distributed query processing is typically performed by either shipping the data required for the query back to the site that issued the query for processing (data shipping), or shipping pieces of the query out to the sites holding the data for parallel processing, returning only the result to the issuing site (query shipping) [14]. These techniques can further be combined as a form of hybrid shipping [10]. Mutant query plans [21] can also provide a form of combined shipping that allows asynchronous query evaluation.…”

“…This has given rise to a rich literature concerning query optimization techniques that allow the system to explore relationally-equivalent representations of a user's query with the end goal of finding the best possible query execution plan (e.g., [14,10]). Traditionally, the term best has been defined in reference to minimizing some combination of overall query execution time or data transmission requirements [14].…”

Robust Placement of Mobile Relational Operators for Large Scale Distributed Query Optimization

“…Hence, the placement of relational operators plays an important role. When we look at the placement methods provided by the parallel and distributed query processing field [6,7,10,11,13,15,23,25,32,41] are based on single-point estimations. Such a placement is as good as compiletime estimations but it yields bad performance in cases of estimation errors.…”