I/O architectures are today an important part of all systems. The two

main types of I/O are network I/O and storage I/O. Recent developments

in underlying technologies and trends in storage systems lead to the

possibility of handling network and storage I/O as a unified

subsystem.



Moreover, there is a trend towards increasing the functionality of the

I/O stack both in network and storage I/O. For instance, modern

IP-based networks, require nodes at the edge of the network to support

functionality related to VPNs, filtering for security purposes,

application specific functions for billing, etc. Similarly, storage

I/O stacks have started to include functions related to storage

virtualization, such as partitioning, aggregation, snapshots,

dual-pathing, compression, encryption, etc. This functionality may

either be placed at the edge or in the core of a system.



Moving functionality to lower system layers in future storage and

network I/O subsystems creates two broad categories of problems: (i)

to satisfy increasing demands in performance and functionality by

designing architectures and building systems that scale in performance

to a wide range of requirements and (ii) to reduce the cost of such

systems. Both of these will lead potentially to new applications, not

possible today, especially related to storing online and transfering

between systems large amounts of information.



The goal of this work is to examine: (i) How CMPs can be used to

perform I/O related processing in modern subsystems. This includes,

understanding what functionality should be supported, how this may be

mapped to generic and commodity architectures, and which are the

aspects that may limit scalability to higher throughput and more

processing capacity. (ii) How such I/O subsystem architectures that

make use of CMPs can potentially be replaced by the application CMPs

in the system, leading to systems with a single CMP that is used both

for application and I/O processing, further reducing cost and

enhancing performance scalability.

The funding for this cluster will be used mostly on exchange of researchers to tighten the relations and help start the collaboration work under the framework of the project. This exchange will be in form of bilateral visits. In addition we plan to organize regular meetings to coordinate the efforts, layout the strategy and decide on the directions of the project. Also we plan on using the funding to attend HiPEAC events such as the Cluster meetings and conferences of interest related to the project.(3 meeting trips x 5 partners to travel x 1.5 people/partner x 1500 euro/trip = 33750 euros)

Paraskevas, EVRIPIDOU, skevos@cs.ucy.ac.cy

Pedro, TRANCOSO, pedro@cs.ucy.ac.cy

Costas, KYRIACOU, cskyriac@cs.ucy.ac.cy

Kyriakos, STAVROU, tsik@cs.ucy.ac.cy

Marcelo, CINTRA, mc@inf.ed.ac.uk