Parallel Computing and Mobile Systems Lab

The Parallel Computing Laboratory (LCP) at COPPE/UFRJ was created in 1985; currently, it is called COMPASSO (Parallel Computing and Mobile Systems). During 1988-93, the COMPASSO team developed the prototype of the first national parallel computer (NCP-I), with a distributed memory architecture. The NCP-I was built with 16 32-bit Transputer microprocessor nodes (Inmos Ltd) and used in subsequent years in parallel engineering applications by several COPPE researchers. This result placed Brazil in the select group that demonstrated technological mastery to develop its own high-performance computing system, fostering its technological independence, as highlighted by the NY Times in August 1990.

Professor Claudio Luis de Amorim presenting the NCP-I. Source: Jornal do Comércio, April 6, 1990.

This text from the NY Times places Brazil, with its NCP-I, among the select group that dominated supercomputer development in the early 1990s. Source: New York Times, August 21, 1990.

In 1991, COPPE , through a contract with the Seismic Exploration Department/PETROBRAS, trained engineers in the nascent parallel processing technology using the NCP I. From 1994 to 2000, COMPASSO developed the NCP 2 parallel computer, which extended the NCP - I architecture with new hardware support for distributed shared memory architecture, and expanded the processing capacity of the NCP 2 with the new T9000 and Intel i860XP chips.

Throughout the years 1999 to 2004, the increasing spread of local area networks and the expansion of the Internet motivated the COMPASSO team to develop new video streaming techniques (patented at the USPTO) Cooperative Video Cache (CVC) and Collapsed CVC , which were implemented and tested in two prototypes of scalable VoD (Video-on-Demand) systems (SMAD 1999-2000, REMAV-RJ) in a communications operator's laboratory with up to 2,000 simulated clients. In the following years (2004/07), efficient, reliable, and portable clusters were built with low-cost servers for several other important classes of web services, in particular a distributed Java platform (cJava) with the Tomcat/JOnAS (Java Open Application Server) standard. The prototypes were tested in real-world environments (SEMSAÇÃO).

Starting in 2005, their team began research, development, and software integration for building scalable VoD distribution systems with access via wireless devices using WiFi technology, meeting defined security and quality of service requirements ( TRAVIS-QoS 2005/07). In the ORLA DIGITAL project (2008/09), in partnership with the RAVEL Laboratory (High-Speed Networks) team at COPPE, the COMPASSO team deployed the WiFi network along the 4.6 km of Copacabana beachfront, where the aforementioned distribution system was successfully tested.

With WEBE (2009/10), R&D began in sensor networks and mobile networks, which are increasingly present in the daily lives of people and companies. This resulted in new techniques (2006/07) with a patent granted by USPTO to UFRJ in 2012. The team also developed the new D1HT (Distributed Hash Table) technique with low latency and small maintenance bandwidth for file systems with more than 1 million pairs, with applications in HPC ( High Performance Computing ) centers and on the Internet. In 2009, COMPASSO proposed and implemented the new REPI protocol for interest-centered ad hoc networks, or RADNETs , in Linux , facilitating the development of applications for ad hoc mobile networks. RadNets can be built with smartphones (Android) and sensor networks with Raspberry Pi (Debian) using Wi- Fi. Currently, the laboratory collaborates with other national research groups on the use of Radnets in applications in the areas of telemedicine, security, precision agriculture, vehicular networks and unmanned aerial vehicles (UAVs), to name a few.

With the RSVP project ( 2009/14), the laboratory leveraged the expertise of its members and external collaborators in the area of HPC and Scalable Web Services to study new Cloud Computing environments, specifically those of IaaS (Infrastructure as a Service). In the context of this project, elastic VoD transmission systems were developed, as well as new timing mechanisms for computing systems based on multicore processors, frequency scaling, and virtualization.

Currently, the laboratory team continues to use its expertise in the HPC field in research focused on Cloud Computing, especially studying the problems of evaluating the capacity and performance of IaaS architectures for scalable applications and the security of these environments.