System architectures
Focus
The term system architecture refers to a quite large research area that includes parallel and distributed architectures as well as network servers down to intelligent embedded systems and pervasive sensor networks. Such a large spectrum of architectures, in continuous evolution, requires facing problems of modeling and design space exploration following flexible approaches that can be adapted to the optimization of different figures of merit, whose relevance changes depending on the application environment. Effective results have been obtained in these years by synergically studying both the architectural characteristics of the single components of the systems, the application properties and the design methodologies oriented toward the concurrent development of hardware and software components.
The group of System Architectures is active on several research topics, covering both foundational and application-driven aspects. A short list is the following: Security, Performance Evaluation, Dependability, Electronic Design Automation, Processor and Multiprocessor Architectures, Operating Systems, Embedded Systems, Computational Intelligence, Wireless Sensor Networks, Enterprise Digital Infrastructure.
The group of System Architectures is active on several research topics, covering both foundational and application-driven aspects. A short list is the following: Security, Performance Evaluation, Dependability, Electronic Design Automation, Processor and Multiprocessor Architectures, Operating Systems, Embedded Systems, Computational Intelligence, Wireless Sensor Networks, Enterprise Digital Infrastructure.
Most relevant research achievements
Self-aware Computing, Adaptive and Reconfigurable Systems
The exploitation of dynamic reconfigurability for FPGA platforms is a recent option and it requires enabling designers to model, validate and evaluate different target reconfigurable architectures. The addressed problems are the high-level modeling, mapping and the definition of feasible architectures for the exploitation of dynamic reconfiguration capabilities (www.dresd.org). Part of this work is being developed for specific architectures in the FP7 FASTER project, additional contributions: Application Heartbeats (a simple, standardized way for monitoring applications performance - http://code.google.com/p/heartbeats/) and AcOS, Autonomic Operating System for Adaptive Computing, (a Self-Adaptive Operating System, starting form the Kernel of the Linux OS - www.changegrp.org/acos).
Power Management and Resource Management for Embedded Computing Systems
CPM (Constrained Power Management) is a formal model and a framework to manage the resources of mobile multimedia embedded systems running a general purpose operating system. The goal is to provide a system-wide framework that is distributed, scalable, with low overhead and capable to identify the optimal trade-off between expected performances (required QoS) and the energy requirements. The entire framework has been developed and tested under Linux and submitted to international patent with STMicroelectronics. Moreover, the group developed also the Barbeque Run-Time Resource Manager, an open-source framework for new generation many-core systems (https://bitbucket.org/bosp).
Methodologies for Analysis, Synthesis and design exploration of Heterogeneous Systems
The main achievement is a framework with algorithms to perform analysis and synthesis for heterogeneous multi-processor systems (PandA) http://trac.elet.polimi.it/panda. In addition a software platform for transaction-level simulation of heterogeneous system architectures has been developed and enhanced with design exploration algorithms for optimal mapping of parallel applications onto multicore architectures and wireless sensor networks.
Exploration and Optimization techniques for Multiprocessor Architectures
The aim is to optimally tune the multiprocessors platform parameters to find the best trade-offs in terms of the selected figures of merit (e.g., energy and delay). The main achievements are a framework for automatic DSE together with advanced algorithms and open source tool (M3Explorer, http://m3explorer.sourceforge.net).
Network on-Chip Architectures, Services and Design Methodologies
The group in the last years developed a framework to support the design-time specialization for NoC paradigm by considering both technological, power and performance dimensions. Moreover, the group designed services for NoC architectures with monitoring and security objectives exploited within STNoC architecture.
Software performance and power analysis
The group has been one of the first in investigating analysis and optimization models for embedded software. This work led to the development of novel software performance and power analysis methodologies and tool-suite. The latest achievements within the context of the EU-FP7 COMPLEX project have led to the implementation of a flexible tool-chain based on the LLVM compilation framework.
Methodologies for the Design of Dependable System
The group proposed a methodology for hardening embedded systems to be implemented on heterogeneous architectures with reconfigurable accelerators. The latest achievements are a framework for the design and implementation at system-level and on (multi-) FPGA platforms and one for the analysis of reliability-related properties.
Intelligent Monitoring Systems
The research addressed methodological and application-related aspects of Intelligent Monitoring Systems encompassing wireless and wireless-wired technologies for environmental monitoring. Power-aware, energy harvesting, QoS, strict-real time and ability to survive in harsh environments are the key issues. Deployments: coral-reef wireless monitoring system, AUS, 2007; rockfall collapse monitoring, (S.Martino Mountain 2010; Torrioni di Rialba 2010, IT) and landslide monitoring (Torrioni di Rialba 2011, IT).
Recommender systems for TV applications
Recommender systems can be used to provide the users with personalized lists of TV programs that they may like. The research group has developed a large number of recommender algorithms tailored for the TV domains. Moreover, the group is internationally recognized for its expertise in the comparative evaluation of recommender systems.
Design and evaluation of security in computer architectures
The group has been active in the areas of efficient design and implementation of cryptosystems, advancements in passive and active side channel analysis methodologies. Some research activities and products have been integrated within STMicroelectronics and HPLabs research and development workflow, resulting in co-authorship of European and International patents: 7 granted, 2 pending, 1 filed.
Emerging threat analysis
We made several contributions in the application of learning algorithms to information security threats. More recently, we designed BURN, an interactive visualization tool for displaying autonomous systems exhibiting rogue activity. We also explored automation of malware analysis, proposing Reanimator and Beagle, two systems that exploit similarities in the code base among different malware samples to identify dormant behaviors. We are currently focusing on emerging threats on mobile devices. Our seminal experiment on Bluetooth malware propagation (codenamed Bluebag) received worldwide media attention. We are now working on automating the analysis of emerging mobile threats.
Analytical techniques and Tools for performance evaluation and optimization Development of new analytical techniques like mean field analysis, asymptotic analysis, markovian agents, for the performance evaluation and optimization of very-large systems, critical infrastructures, sensor networks, hazard propagation. Moreover, we developed new features of the Java Modelling Tools suite JMT: fork/join for parallel computing, traffic traces fitting, bottlenecks identification with two and three classes of requests, regions with capacity limitations.
Datacenter Consolidation and Green IT
Power and cost reduction in large-scale datacenters can be achieved by adopting server consolidation strategies based on using optimization models able to consider different sets of performance and dependability requirements, and by considering cooling-aware workload placement that takes into account the temperature of the cold air supplied by the air conditioner. This allows to reduce the power consumption of the servers, while satisfying a number of constraints concerning temperature and response times.
Incremental functional diagnosis
Work in this area has been devoted to investigating new directions for the diagnosis of complex systems, focusing on efficiency and confidence in the results, while trying to minimize the amount of effort and tests. The main achievement is a methodology for performing adaptive, incremental functional diagnosis to reduce costs but keeping a high accuracy.
The exploitation of dynamic reconfigurability for FPGA platforms is a recent option and it requires enabling designers to model, validate and evaluate different target reconfigurable architectures. The addressed problems are the high-level modeling, mapping and the definition of feasible architectures for the exploitation of dynamic reconfiguration capabilities (www.dresd.org). Part of this work is being developed for specific architectures in the FP7 FASTER project, additional contributions: Application Heartbeats (a simple, standardized way for monitoring applications performance - http://code.google.com/p/heartbeats/) and AcOS, Autonomic Operating System for Adaptive Computing, (a Self-Adaptive Operating System, starting form the Kernel of the Linux OS - www.changegrp.org/acos).
Power Management and Resource Management for Embedded Computing Systems
CPM (Constrained Power Management) is a formal model and a framework to manage the resources of mobile multimedia embedded systems running a general purpose operating system. The goal is to provide a system-wide framework that is distributed, scalable, with low overhead and capable to identify the optimal trade-off between expected performances (required QoS) and the energy requirements. The entire framework has been developed and tested under Linux and submitted to international patent with STMicroelectronics. Moreover, the group developed also the Barbeque Run-Time Resource Manager, an open-source framework for new generation many-core systems (https://bitbucket.org/bosp).
Methodologies for Analysis, Synthesis and design exploration of Heterogeneous Systems
The main achievement is a framework with algorithms to perform analysis and synthesis for heterogeneous multi-processor systems (PandA) http://trac.elet.polimi.it/panda. In addition a software platform for transaction-level simulation of heterogeneous system architectures has been developed and enhanced with design exploration algorithms for optimal mapping of parallel applications onto multicore architectures and wireless sensor networks.
Exploration and Optimization techniques for Multiprocessor Architectures
The aim is to optimally tune the multiprocessors platform parameters to find the best trade-offs in terms of the selected figures of merit (e.g., energy and delay). The main achievements are a framework for automatic DSE together with advanced algorithms and open source tool (M3Explorer, http://m3explorer.sourceforge.net).
Network on-Chip Architectures, Services and Design Methodologies
The group in the last years developed a framework to support the design-time specialization for NoC paradigm by considering both technological, power and performance dimensions. Moreover, the group designed services for NoC architectures with monitoring and security objectives exploited within STNoC architecture.
Software performance and power analysis
The group has been one of the first in investigating analysis and optimization models for embedded software. This work led to the development of novel software performance and power analysis methodologies and tool-suite. The latest achievements within the context of the EU-FP7 COMPLEX project have led to the implementation of a flexible tool-chain based on the LLVM compilation framework.
Methodologies for the Design of Dependable System
The group proposed a methodology for hardening embedded systems to be implemented on heterogeneous architectures with reconfigurable accelerators. The latest achievements are a framework for the design and implementation at system-level and on (multi-) FPGA platforms and one for the analysis of reliability-related properties.
Intelligent Monitoring Systems
The research addressed methodological and application-related aspects of Intelligent Monitoring Systems encompassing wireless and wireless-wired technologies for environmental monitoring. Power-aware, energy harvesting, QoS, strict-real time and ability to survive in harsh environments are the key issues. Deployments: coral-reef wireless monitoring system, AUS, 2007; rockfall collapse monitoring, (S.Martino Mountain 2010; Torrioni di Rialba 2010, IT) and landslide monitoring (Torrioni di Rialba 2011, IT).
Recommender systems for TV applications
Recommender systems can be used to provide the users with personalized lists of TV programs that they may like. The research group has developed a large number of recommender algorithms tailored for the TV domains. Moreover, the group is internationally recognized for its expertise in the comparative evaluation of recommender systems.
Design and evaluation of security in computer architectures
The group has been active in the areas of efficient design and implementation of cryptosystems, advancements in passive and active side channel analysis methodologies. Some research activities and products have been integrated within STMicroelectronics and HPLabs research and development workflow, resulting in co-authorship of European and International patents: 7 granted, 2 pending, 1 filed.
Emerging threat analysis
We made several contributions in the application of learning algorithms to information security threats. More recently, we designed BURN, an interactive visualization tool for displaying autonomous systems exhibiting rogue activity. We also explored automation of malware analysis, proposing Reanimator and Beagle, two systems that exploit similarities in the code base among different malware samples to identify dormant behaviors. We are currently focusing on emerging threats on mobile devices. Our seminal experiment on Bluetooth malware propagation (codenamed Bluebag) received worldwide media attention. We are now working on automating the analysis of emerging mobile threats.
Analytical techniques and Tools for performance evaluation and optimization Development of new analytical techniques like mean field analysis, asymptotic analysis, markovian agents, for the performance evaluation and optimization of very-large systems, critical infrastructures, sensor networks, hazard propagation. Moreover, we developed new features of the Java Modelling Tools suite JMT: fork/join for parallel computing, traffic traces fitting, bottlenecks identification with two and three classes of requests, regions with capacity limitations.
Datacenter Consolidation and Green IT
Power and cost reduction in large-scale datacenters can be achieved by adopting server consolidation strategies based on using optimization models able to consider different sets of performance and dependability requirements, and by considering cooling-aware workload placement that takes into account the temperature of the cold air supplied by the air conditioner. This allows to reduce the power consumption of the servers, while satisfying a number of constraints concerning temperature and response times.
Incremental functional diagnosis
Work in this area has been devoted to investigating new directions for the diagnosis of complex systems, focusing on efficiency and confidence in the results, while trying to minimize the amount of effort and tests. The main achievement is a methodology for performing adaptive, incremental functional diagnosis to reduce costs but keeping a high accuracy.
