Research and Development Line
1. Modeling and Auto tune of “The Machine” performance
In this research line, we will provide a model to capture The Machine characteristics, focusing on the effects that the universal memory and the photons network cause in the application executions. We will investigate the ways to automatically find the best parameters of compilation to adjust (“tune”) the applications to The Machine.
The main idea is deriving a theoretical model in which many possibilities of scheduling algorithms or even complexity results can be investigated.
We have been working on techniques and frameworks of empiric self-adjustment (“auto-tuning”), such as opentuner.org, to optimize the search for algorithmic choices e combinations which can solve a given problem. The search spaces generated by this type of problem are enormous.
Motivated by recent advances in computer architecture and its wide variety, we propose to applicate empiric self-adjustment techniques in problems of automatic parallelization and compiler optimization in different processor architecture. We are also interested in do all this in a simulated way, possibly in a cloud computing environment, since we are not able to predict the initial availability of many The Machine.
1.1 Members and Research
Pedro Bruel
The main motivation of our research on autotuning for The Machine is to find ways to leverage the increasing heterogeneity of parallel and distributed programming models and architectures. The research involves the study and application of search heuristics to optimize programs and configurations for hardware accelerators that have the potential to be part of The Machine.
Pedro applies search heuristics and develops tools for autotuning programs and configurations for CPUs, GPUs and FPGAs. He studies the applicability and the performance of existing tools, such as the OpenTuner framework (http://opentuner.org/), in different domains and architectures for High Performance Computing. Pedro also develops OpenTuner extensions for cloud execution, and an autotuning library in the Julia language (https://github.com/phrb/StochasticSearch.jl).
2. Modeling and Simulation of Smart Cities
In short, this research line has as long-term objectives the following:
· To clarify requiments for smart cities
· To modulate concrete application scenarios (different application classes and different situations in cities)
· To develop a simulator (free software ) for smart cities
· To propose advanced software engineering tools to the development of smart cities applications
· To integrate software engineering tools and the simulator in only one environment
· To conceive and to develop a App Store for The Machine
2.1 Members and Research
Arthur Del Esposte
Arthur conducts research on key technical issues related to the development and deployment of Smart Cities platforms. The first part of the research involves evaluating and comparing existing Smart Cities platform via simulations and technical studies. The complementary part of the research aims to develop a scalable, integrated Smart City platform prototype to address technical open issues in the area. The ongoing development platform is based on a microservices architecture and is open source available at: https://gitlab.com/smart-city-software-platform.
Smart Cities platform are applications that can potentially benefit from TheMachine's architectural innovations to meet the key requirements of scalability, efficiency and data management. In particular, the research line developed by Arthur aims to identify key technical challenges that should be considered in Smart Cities platform architectures to benefit from advances provided by TheMachine. The under development Smart Cities platform may be continuously evaluated through experiments guided by realistic scenarios of Smart Cities, primarily supported by Smart Cities simulators studies.
Eduardo Santana
Eduardo is responsible for the development of SCSimulator, a smart city simulator that aims to help tests of smart city applications and platforms and help the decision-making of city managers about various city investments such as the position of sensors in the city IoT network, the effects of a intelligent traffic light system, and the impact of a change of a bus line.
The simulator is important in the The Machine project because it is an application that need of a great computation power, because to simulate a big city, such as Sâo Paulo, will be necessary the execution and management of millions of actors such as cars, sensors, traffic lights, and pedestrians.
Rafael Aquino
Big Data frameworks have as main objective to process a big amount of data, and these processes cannot be done in common systems. This research have the goal to compare big data frameworks, considering the execution time as the main parameter. Our main framework choices for the comparison are HPAT (High Performance Analytics Toolkit) and Spark, because both use memory as their main storage during the execution, and this brings more efficiency.
Our main objective is to find which framework works better in a memory-based environment, and which advantages and disadvantages were found in these frameworks during the research. With these informations, we can try to predict which of these frameworks will work better in an environment like the The Machine.
Rodrigo Tinini
Optical Networks are a communication technology that permit high rates of transfer and speed. The objective of this research is to propose new algorithms and protocols to the operation and management of optical networks in two different scenarios: Supporting the transmission between pairs of an optical network in traditional long distance topologies and to interconnect processors in multicore systems on-chip, which is called Optical Network on-Chip, providing high speed and high transfer rates among hundreds or even thousands of processors on a single chip.
Our objective is focused on the optimization of the operation of optical networks regarding the increasing amount of data generated nowadays and on important issues like the overall energy consumption of these networks. Mainly, we focus on developing new Routing and Wavelength Assingment - RWA and Routing, Modulation and Spectrum Allocation - RMSA algorithms for WDM and Elastic Optical Networks, respectively, and also optimization models based on Integer Linear Programming.
