Muhammet Orazow
Data locality aware partitioning schemes for large-scale data
stores
Tese submetida para provas de mestrado em Engenharia Informática e de Computadores
Instituto Superior Técnico, Universidade Técnica de Lisboa.
Abstract
Key-value stores are highly scalable storage
systems that can offer extremely good performance. For these reasons
key-value stores are the backbone of several large-scale data
processing systems. However, their interface is rather restrictive
since it only allows to access objects through their keys. To overcome
this limitation, recently proposed systems have developed mechanisms
for storing data in multiple dimensions mappings in order to allow
searching for objects in using their secondary attributes. These
solutions, however, pose another serious problem: that of configuring
the system such that it may take the best advantage of these
multi-dimensional mappings.
This thesis makes two main contributions towards the automatic
configuration of such multi-dimensional key-value stores: First, from
a detailed description of the inner workings of these systems, we
derive a model which describes the behavior of queries in
multi-dimensional spaces. We then use this model to predict real
throughputs of the system for complex workloads. Using these results,
we propose a generic architecture which allows to automatically adapt
the configuration of the multiple dimensions, in order to obtain the
maximum possible throughput for a given workload.
Publicações
- Data locality aware partitioning schemes for large-scale data stores
- Muhammet Orazow
- MSc Thesis. Instituto Superior Técnico,
Universidade Técnica de Lisboa.
- July, 2013.
- Available BibTeX, MSC Thesis, and extended abstract of the
thesis.
- Autonomic Configuration of HyperDex via Analytical
Modelling.
- N. Diegues, M. Orazov, J. Paiva,
L. Rodrigues and P. Romano.
- The 29th Symposium On Applied
Computing (SAC 2014), Gyeongju, Korea March, 2014
-
- Auto-Configuração de Bases de dados
NoSQL Multi-Dimensionais.
- N. Diegues, M. Orazov,
J. Paiva, L. Rodrigues and P. Romano.
- Actas do quinto
Simpósio de Informática (Inforum), Évora,
Portugal, Sep, 2013
- Available BibTeX, extended report
(pdf).
Luís Rodrigues