Research Projects

PC Clusters have become the most attractive platform to implement supercomputers and high-performance servers. As a consequence, it is necessary to improve these platforms in order to provide higher performance and scalability, and better services, such as availability, reliability, quality of service, low power, etc. It is also necessary to improve the access to servers through Internet. In this project, we propose to improve most of the hardware components of a PC cluster. We will also improve the cluster system software. In particular, we propose to improve processor performance by using superpipelining techniques and by reducing power consumption. I/O system performance will be improved by using a switch fabric with point-to-point links. SMP scalability will be improved with a new memory coherence scheme. With respect to the interconnection network, we propose improving not only its performance (by using adaptive routing techniques) but also its scalability (by using load balancing techniques), availability (by using fault-tolerance techniques) and quality of service (by including some hardware/software support in the routers and network interface controllers). We will also develop scheduling techniques that will take into account communication costs. Regarding Internet access, we will also enhance IP routers, improving their scalability and bandwidth by means of new congestion control mechanisms and HOL blocking avoidance techniques. We will also research on wireless networks, developing new robust coding schemes based on new UEP techniques and applying them to the transmission of MPEG-4 video. We will also develop some distributed applications focusing on those that require some specific hardware to allow real-time execution. In particular, we will develop quantization and coding procedures and their hardware implementation on FPGAs. Finally, we will improve the synthesis of FPGA circuits from highlevel specifications.

Since the beginning of the WWW, not much technologies have attracted so enthusiastic attention from a wide range of different users of the net than the Grid Technology. This technology, that offers the support for a highly distributed computation, is based in three main pillars:

• Network infrastructure based on IP technology Authentication using public key infrastructure based on X.509 Software support and tools to write distributed applications within this context.
• Over these three pillars, many projects have started to provide Service-Oriented Architectures (SOA) so that users can access all the services offered by the Grid in an easy way. Over all of them, the Globus Project has become the de-facto standard.

However, although there are a lot of research in the area of Grid Computing, this technology faces new challenges due to the evolution of IT (Information Technologies). For example, two of the most important challenges, that have been identified by the experience of our research group, and that are the main research areas of our project are:

• Adapting the network infrastructure and the security services, and the public key infrastructure of the Grid to the evolution from IPv4 to IPv6.
• Addition of new services of distributed computation, such as those based on Component-Based Development (CBD), component models and integration with Web Services.

As can be seen from the project goals, the project approach is twofold: from the network infrastructure (network protocols, communication), and from a service and component-oriented perspective, so that we can extend a platform such as Globus with these new services.

The PITAGORAS project tries to develop a software infrastructure (middleware) for building distributed applications in network of PCs (NOWs). This middleware offers metacomputing services, such as resource management, transparent process and data migration, replication, fault tolerance, intelligent load balancing, QoS (Quality of Service), Security, etc. Special care will be taken in designing a compact and efficient software, robust and easy to use, extensible. Moreover, the developed software should be easily installable, based on standard tools and protocols, such as the CORBA standards. Using CORBA allows us to take advantage of other projects developed using CORBA, and also to offer a clean interface to other CORBA developments. The project goal includes defining a well-stablished interface so that any application can use the services offered.

Networks of workstations and clusters of personal computers (PCs) have become the most attractive platform to implement supercomputers, high-performance servers, and distributed systems. Therefore, these platforms need to be continuously improved in order to offer higher performance as well as better services, including higher availability and reliability, quality of service, low power consumption, scalability, etc.

In this project we propose several research activities to improve all the hardware components of a cluster of PCs, including the associated system software. In particular, we propose improving processors by simplifying processor hardware and using dynamic compiling techniques, improving the I/O system by using networks with point-to-point links, improving the scalability of symmetric multiprocessors by means of excess directories, and improving the interconnection network. With respect to the network, we propose improving performance (by means of adaptive routing) as well as scalability (by means of load distribution techniques), availability (by means of dynamic reconfiguration), and quality of service (by means of hardware or software support). We will also study the use of wireless networks in distributed systems where GSM is not appropriate. In particular, we will use them to develop very large sensor arrays. Finally, we will develop several distributed applications, focusing on those that require hardware support for real-time execution. In particular, we will develop video compressors/decompressors based on wavelet transform and real-time artificial perception techniques applied to teledetection and several engineering fields. For these applications, we will develop hardware accelerators based on FPGAs.

The main goals of this project are the design of implementation of high-performance low-cost network of workstations, and the development of parallel applications for computer vision and image processing. These applications will also be useful to evaluate the performance of proposed network of workstations. We plan to build a prototype machine using personal computers and evaluating several options for the interconnection network, from Fast Ethernet cards to custom designs based on FPGAs. We will develop and evaluate new switching techniques and routing algorithms that matched the specific requirements of networks of workstations. We will focus on achieving high perfomance and reliability as well as low cost. We will also develop parallel applications for computer vision and image processing with the goal of satisfying the computer power requirements of real time processing. These applications will also be used to evalutate the prototype. Finally, it is worth to mention that the proposed networks of workstation has many potential applications, taken into account the increase demand of parallel and distributed applications (distributed data-bases, multimedia applications with real-time video transmission, etc), which require high computing power and efficient communications.