ICS-FORTH 25-Nov Infrastructure for Scalable Services Are we Ready Yet? Angelos Bilas Institute of Computer Science (ICS) Foundation of Research and Technology – Hellas (FORTH)
ICS-FORTH 25-Nov Outline Who we are & what we do Issues with future computing infrastructure Topics of interest within DeSIRE/DeFINE
ICS-FORTH 25-Nov Foundation Of Research and Technology- Hellas (FORTH) Largest Research Center in Greece –Located in Heraklion, Crete 1000 Employees –100 University Professors –120 Researchers 34M€ Annual Budget –20% Government –260 Research Projects 6 Spin-off Companies
ICS-FORTH 25-Nov Institute of Computer Science (ICS) One of seven institutes in FORTH –20 Researchers –50 Staff Members –120 Students Cutting-edge research in various areas –Information Systems/Databases –Computer Systems/Networks –Human Computer Interaction –Vision and Robotics –Medical Informatics
ICS-FORTH 25-Nov Computer Systems Research Topics Systems software Middleware and runtime systems Parallel and distributed systems Interconnection networks Application specific microprocessors VLSI and hardware design Applications
ICS-FORTH 25-Nov Access Points Computational View of the World Model for numerous services –Thin access points (clients) –Scalable backend servers –Connectivity Today, lot’s of pressure on backend –More and more access points Continuously increases –Existing services are offered directly to customers, e.g. medical info –New services are made possible, e.g. entertainment industry and multiplayer games Services Network
ICS-FORTH 25-Nov State-of-The-Art for Scalable Infrastructure There has been a lot of progress on performance However, many remaining problems –Manageability e.g. Dynamically/automatically adjust to environment changes –Ease-of-use e.g. Develop new applications/services –Reliability and Availability –Security
ICS-FORTH 25-Nov Outline Who we are & what we do Issues with future computing infrastructure Topics of interest within DeSIRE/DeFINE –Reliability and Availability –Security
ICS-FORTH 25-Nov Reliability and Availability We do poorly today in scalable, commodity infrastructure –Frequent failures –Slow recovery –Unacceptable situation Recent activity in dealing with problems in the systems community –Conferences: ISCA, ASPLOS, HPCA Most work assumes –Commercial applications (although difficulties to find workloads) –Treat as multithreaded programs –Run on shared memory multiprocessors
ICS-FORTH 25-Nov Our Approach [HPCA2003] Use commodity clusters as the underlying system Run multithreaded applications without modification –Software shared memory Provide reliability support in middleware –Runtime system Dynamic replication of system and application state at runtime When failure occurs, use the second copy to recover Any application Interconnect System Abstraction (Middleware)
ICS-FORTH 25-Nov Lessons Learned Design and implementation on a 64 processor cluster –It works –Performance impact can be small Between % of execution time –Transparent to applications –Replication is a powerful mechanism But… –Very complex: Mostly due to implementation issues –I/O semantics lead to unclear failure semantics –Don’t cheat with applications: Use commercial apps
ICS-FORTH 25-Nov Going Forward Need to understand –Nature of failures –Interfaces between system layers –State in each system layer –Application requirements (but strive for transparency) Design and build –Realistic prototypes that demonstrate solutions –In real setups with real applications Bridge gab between systems building with more rigorous approaches
ICS-FORTH 25-Nov Security (contact person Prof. Evangelos Markatos) Security is becoming an inhibiting concern –A lot of focus on improving throughput/response times –However, various types of attacks that cost a lot Difficult problem –Fast networks, advanced features (QoS, etc.) Our approach: Use the network interface –Work at lower system layers, closer to the network –Take advantage of technology trends (network processors, etc.) Goal: Examine how we can integrate methods for –Intrusion detection –Denial of service attacks Deliver hardware/software to demonstrate realistic solutions
ICS-FORTH 25-Nov Summary Solid systems expertise Domain –Scalable computing infrastructure for novel services Topics of interest within DeSIRE/DeFINE –Reliability and Availability –Security Welcome collaborations with groups that have –Similar interests –Complementary expertise Draft a document for work in the area
ICS-FORTH 25-Nov Contact Information Angelos Bilas Associate Professor Institute of Computer Science (ICS) Foundation of Research and Technology – Hellas (FORTH)