IOLanes is a EU-funded research project targeted at understanding and improving the I/O performance in modern hardware that employees multicore architectures by adapting or redesigning the I/O stack, as appropriate and by providing system-level support that will allow future storage systems to take advantage of multicore CPUs in new ways.
Storing and processing data, e.g. large indices, web content, telephone logs, profiles, purchase transactions has emerged as a main problem in modern ICT infrastructures. Projecting forward, emerging applications and services will require both processing as well as will generate an unprecedented volume of data and information. However, today's infrastructures rely on I/O subsystems that are not able to cope with such requirements and the performance of the I/O subsystem has been, in many cases, limiting how ICT infrastructures are provisioned.
With the continuing increase in the number of cores, processing power and hardware speed, contemporary hardware can perform a vast amount of I/O operations per second and achieve very high I/O throughputs and low latencies. For instance, solid-state disks (SSDs) and system interconnects are capable of throughputs in the range of 10 GBytes/s and I/O operations in the range of hundreds of thousands to millions. However, typical servers today, are only able to sustain an order of magnitude lower I/O performance. A main obstacle to realizing the potential of modern architectures is systems software that has a hard time catching up with advances in hardware, and, more often than not, is the bottleneck in modern systems. To benefit from the increasing amounts of disk storage and network bandwidth, all software layers have to be examined and rethought.
In addition, today's I/O subsystems are complex, multi-layer systems that are hard to understand and extend with new functionality. For instance, extending the I/O subsystem in an existing installation to support an alternative I/O scheduler for a class of workloads, becomes a daunting task. Similarly, adding functions, such as compression for improving space efficiency or performance (and in both cases ultimately system cost) or encryption for increased access control requires extensive effort. Our I/O subsystems have not been built to support such requirements. A key to addressing these issues is taking advantage of emerging multicore CPUs that offer ample compute cycles to offer generic support for I/O extensions.
IOLanes follows a vertical, cross-cutting approach to address these issues and brings together expertise from I/O platforms, applications, virtualization, performance, and architectures to design and prototype a scalable I/O stack for modern storage infrastructures. We expect that IOLANES will achieve unprecedented I/O rates at high cost-efficiency. This will allow applications and services to both scale to larger data sizes and customer bases as well as lower the cost barrier to entry for small businesses and cost-sensitive markets.
The IOLanes project is supported by partners Foundation for Research and Technology - Hellas (FORTH) - Coordinator, Barcelona Supercomputing Center (BSC), University of Madrid (UPM), INTEL Performance Labs Ireland, IBM Research Labs in Haifa, and Neurocom Ltd. IOLANES is funded by the EC under the 7th Framework program and is part of the portfolio of the Embedded Systems Unit - G3 Directorate General Information Society.
