Dominique Verchere
Dr Dominique Verchere
Bell Labs France
Download presentation
Scheduling, Reconfiguration & Virtualization service: the introduction path for grid networking
Amongst the drivers for technological evolution of high-capacity transmission networks, the advanced functional requirements for performance of the connection services delivered to large-scale distributed applications (referred as Grid Applications in Service oriented Architecture 1) play a key role.
The sporadic data traffic generated by Grid applications requires specific QoS-constrained connectivity parameters. For instance, ultra-high bandwidth for massive data transfers, low latency for real time coordinated measurement equipments (e.g. collaborative radio telescopes to create a single coherent instrument), or specific loss rates for interactive applications displayed on remote visualization picture walls.
The recent advances in optical and switching systems (e.g. Wavelength Selective Switches, agnostic Switching Fabrics) enable to configure large capacity and deterministic network services (e.g. low BER and no jitter). However, Grid applications are currently overlaid on the transport networks, precluding the consistent and dynamic configurations of the connection services per grid application workflows. Interfacing network management functions with grid application management functions are the new challenges to overcome the limitation of the overlay model.
In the paper we develop how the optical transport network infrastructure can intercept and deduce from the Grid Application its QoS parameters to set-up, at the right time, the connection services with guarantees in terms of bandwidth, acceptable end-to-end delays and packet loss during the necessary time interval.
Two options of interactions between the grid applications and the networks are presented. First level presents an interface from the network management system to expose network services accessible to the Grid Service Providers (GSP) offering services to the application users. And second level defines protocol gateways for the routing and signaling control functions (based on GMPLS) to enable to configure the grid + network services more directly and uniformly with the relevant amount of resources needed, this during the lifetime of the application workflows.
These two levels of interactions allow the publication, the discovery and the selection of the services taking into account the network capabilities (traffic intensity, QoS parameters, availability) in order to optimally manage the network resources for a global network service responding to the grid application needs.
This challenge can be achieved by providing a new level of virtualization of resources for configuring the network services with information from the grid application servers (computational, storage, visualization or acquisition). The signaling engines should facilitate the reservations including IP/MPLS, Ethernet/T-MPLS, SONET/SDH, and DWDM and their combinations through multiple domains. GMPLS provides network resource virtualization for (re)configuring the network services for the grid application users. The associated scheduling functions perform the arbitration of the different types of the resources spanning different network domains.
Achieving these new capabilities imposes to extend the protocols of the transport network. They have to be designed by considering: heterogeneous resource types and the scalability related to the number of sessions, combination of network resource types, and the complexity of the connections (from point-to-point to multipoint-to-multipoint).
Scheduling and virtualization service to allocate optimally resources for large-scale distributed applications over self-reconfigurable optical networks enables virtual organizations (including Research Laboratories, Private Companies, Public Institutions) to interact/transact with their partners, their customers and their suppliers.
Finally the paper opens on the next challenges to standardize such Scheduling, Reconfiguration & Virtualization (SRV) services taking into account both the NREN and Carrier Networks contexts.
Biography
Dominique Verchère received Ph.D. in Computer Sciences. Since 1998 he has been with Alcatel Research, working first on router control software, data path processing, terabit-switching capacity systems. He worked on resilience for Optical X-Connect and Carrier Grade Ethernet systems with functions according to GMPLS. He contributed in European projects: EuroNGI, TBones, VIOLA and NOBEL1 & 2. He is leading the project on Architectures and Protocols for Computing services offered by Optical Networks (CARRIOCAS). He has published more than 40 papers and 20 patents. He is Distinguished Member of Alcatel-Lucent Technical Academy.