Switching architectures

The activities on switching architectures are a long-lasting research effort of the TNG group. The most active research areas today include: software routers, power-aware scheduling in Input Queued (IQ) switches, disaster recovery, virtual machine migration and performance measurements for cloud computing.

  • Software routers. Software routers based on Personal Computer (PC) architectures are becoming an important alternative to proprietary expensive network devices. However, software routers suffer from some limitations such as reduced number of available interfaces, reduced reliability, limited performance, etc. To overcome these drawbacks, we defined a multi-stage architecture based on the interconnection of PCs to permit increasing the performance of single software routers, to scale router size, to distribute packet-forwarding and control functionalities, to recover from single-component failures, and to incrementally upgrade router performance. A control protocol was defined to manage the multi-stage architecture. We have introduced energy-saving features to switch on/off elements of the multi-stage router based on the traffic load, exploiting the Wake-On-LAN technology combined with traffic prediction and/or on-line measurements.
  • Scheduling in IQ (Input Queued) switches. Scheduling in IQ switches implies controlling input queues to avoid contention when transferring data to output ports. We are currently studying power-aware scheduling algorithms. Instead of looking only at performance maximization, we also try to reduce energy consumption (and power dissipation) by exploiting Dynamic Voltage and Frequency Scaling (DVFS) capabilities in the switching fabric.
  •  Business continuity and disaster recovery. Storage Area Networks (SANs) are mostly based on specific networks and protocols like Fiber Channel or InfiniBand, due to the their high performance and reliability. However, today, thanks to the availability of the iSCSI protocol, IP-based SANs are becoming more appealing because of the ever increasing diffusion of high-speed IP infrastructures, mostly based on Ethernet technologies like 1/10 Gigabit Ethernet. We have evaluated PCs equipped with 10 Gbit/s Ethernet adapters: configuration, throughput and delay measurements, routing capabilities realization of a metropolitan test-bed in collaboration with the regional research centers CSI-Piemonte, CSP and TOP-IX over a 1/10Gbit/s MAN are the main objectives of the research work. Furthermore, Disaster Recovery capabilities are increasingly becoming important design targets for future networks. An important asset to guarantee data reliability and accessibility is the geographical distribution of storage resources exploiting high-speed data networks and remote storage solutions that are becoming commercially available. The distribution of storage resources permits data virtualization, effectively decoupling the data and the service access from the physical devices. Furthermore, the full migration of application services is also available thanks to applications virtualization solutions (e.g. Xen, WMware, OpenVZ, etc.). We are currently performing analysis of virtualization solutions (including modeling, simulation and experiments) and designing a network infrastructure suitable for data and application services migration, to guarantee reliability, flexibility and improve performance.

Contact person: BIANCO ANDREA

Research groups