Nuno Preguiça Cloud Support for Global Computing

Cloud Support for Global Computing

COMPUTER SYSTEMS / CR4 Team

DEPARTAMENTO DE INFORMÁTICA

Photo

Nuno Preguiça Assistant Professor at Dep. Informática – FCT-UNL

Currently working on cloud-based solutions for global computing

Funding: PEst-OE/EEI/UI0527/2011 PTDC/EIA-EIA/108963/2008 (RepComp)  PTDC/EIA-EIA/113613/2009 (Synergy-VM) PTDC/EEI-SCR/1837/2012 (SwiftComp) ANR-10-BLAN 0208 (Concordant)  

Global Internet Services require:

Permanent, fast and safe access to data;

Fast computation over large amount of changing data.

The goal of our research is to:

Develop data management solutions with high availability and low latency;

Develop algorithms for incremental data processing;

Develop solutions for improving efficiency and reliability of multi-core nodes.  

Experimental systems research

Derive requirements and evaluate solutions using realistic workloads

Propose algorithms and system design that are realized in working prototypes

Main approaches leverage operations properties (e.g. commutativity) to reduce coordination requirements and increase load parallelization  

Design of systems that include algorithms for:

Replicated data management for cloud computing with high availability and low latency;

Efficient incremental processing of information;

Replication of software component for improved performance and reliability.

•  The increasing complexity of software has led developers to (re)use third party libraries and components

•  Such components contribute to an increase in software bugs

•  Macro-Components provide an easy to deploy mechanism for detecting and preventing bug occurrences

•  By simply replacing standard components with their Macro-Component siblings

•  Provide increased fault-tolerance to everyday applications with minimum development time and cost

•  Macro-Components are software components that provide increased fault-tolerance

•  Include diverse component replicas of the same specification

•  Leverage on the computational power of multicore processors •  Concurrently executing methods on all replicas

•  Faults are detected by comparing results and states of the internal replicas, thus identifying possible divergences

•  Majority defines the correct result/state •  Diverging replicas are marked for recovery based on healthy ones

•  Offer practical framework for developing Macro-Components •  Easy deployment of fault-tolerant components by simply replacing

standard components with their Macro-Component siblings •  Provide increased fault tolerance without compromising application

performance •  Macro-Components can also offer performance improvements, over

their standard siblings, when configured with homogeneous replicas •  Replicated software components for improved performance. In: Proc. InForum'10 •  Improving Application Fault-Tolerance with Diverse Component Replication. In WTM, 2012 •  Scaling Database Engines on Multicores. Submitted to Euro-Par 2013

Replicated Components COMPUTER SYSTEMS

DEPARTAMENTO DE INFORMÁTICA

Photo

João Soares (PhD Student)

Currently working on RepComp project Advisor: Prof. Dr. Nuno Preguiça

Funding:

!"#$%&'($)*$+,) *$+-) *$+")

./01/2('#3%)

4$#53/6'77)

!"#$%&'($)*$+72(',) *$+72('-) *$+72('")

4$#53/*$#1%")

Research Projects: PTDC/EIA-EIA/108963/2008 Research Grants: SFRH/BD/62306/2009