Master's Thesis

Admission Control for OpenSM/InfiniBand

The OpenSM/OFED stack has only recently (March 2008) gained support for QoS, but without any support for bandwidth limiting/admission control. Previous research has shown that admission control is necessary in order to ensure low latency operation when network load is high. Therefore it is demand for a admission control scheme to supplement the existing OpenSM/OFED/InfiniBand QoS features.

Carto - Topology Aware Processor Allocation with MPI

Carto is a framework for topology aware allocation of processors with OpenMPI. Currently Carto uses the physical topology as a basis for processor allocation in order to improve network performance. As there may be large differences between the physical topology and the logical topology as defined by the routing algorithm, performance can possible be further improved by making Carto routing aware, i.e. exploit the current routing to better utilise the network by proper placement of jobs.

Data Centre Bridging - The Next Step in the Ethernet Evolution

In the networking industry there is currently an initiative for improving the capabilities of Ethernet in the context of data centres. The purpose is to make it possible to converge LAN, SAN, and HPC applications onto a single interconnection network, i.e. Data Centre Ethernet. This is also considered necessary for Ethernet to remain competitive with InfiniBand in high performance computing and advanced data centres. There is an ongoing standardisation effort that seeks to enhance Ethernet with better link level flow-control, traffic classes, congestion management etc. But exactly how to do this is still being researched.

Calibration of Network Simulations Against Hardware Implementations

Simulation represents the most flexible and available method for modelling interconnection networks. E.g. when simulating large InfiniBand clusters, that are infeasible to build for the sake of testing, simulation is a valuable tool to predict the performance characteristics of the cluster. In order to strengthen our confidence in the simulation results it is useful to calibrate the simulation model with a small prototype of its real world equivalent, when such a prototype is feasible.

Exploiting HPC for extreme scaling and performance of video signal processing

Did you ever wanted to make the 3D holograms from Star Wars movies real? Here is your opportunity: Capturing, encoding and displaying natural 3D scenes in real-time requires large camera arrays and massive processing power. At Simula Research we are currently building a massively-parallel framework for 3D multiview video-signal processing. To achieve ultra-low latency, the processing framework may take advantage of the high-speed and low-latency interconnects in a new InfiniBand cluster installed at our laboratory.