GMPLS-based service differentiation for scalable QoS support in all-optical Grid applications

Abstract

In the forthcoming new era of truly distributed computing, industry, businesses, and home users are placing complex and challenging demands on the transport network, now powered by the emerging photonic technologies, about Quality-of-Service (QoS) assurances that are required for new real-time computing and storage service applications geographically distributed worldwide according to the Grid model. There is the need to devise mechanisms for QoS provisioning in IP over WDM networks that must consider the physical characteristics and limitations of the optical domain to ensure the proper treatment of service classes when passing from the electrical switching to the optical domain and back. In addition, these mechanisms should be directly accessible to Grid applications to make them able to request and release network resources as they need. A (G)MPLS-based control plane combined with a wavelength-routed optical network is seen as a very promising approach for the realization of transport infrastructures for the future “photonic empowered” Grid computing paradigm, since it allows native user-controlled bandwidth resources and class-of-service provisioning, that is one of the strongest requirements for truly distributed computing. Considering this, we propose a general framework for providing differentiated services QoS to Grid applications in wavelength-routed photonic networks, built on the strengths of GMPLS for dynamic path selection and wavelength assignment. This framework makes it technically and economically viable to think of a set of computing, storage or combined computing storage nodes coupled through a high-performance optical network as one large computational and storage device.

Introduction

Over the past years, it has become evident to the technical community that computational resources cannot keep up with the demands generated by some processing and data storage bound applications. Furthermore, the Internet, with the IP protocol as the most predominant networking technology, is evolving from best-effort service toward a differentiated service framework with QoS assurances which will be necessary for these applications, such as real-time particle physics experiments or radio astronomical image processing that produce more data than can be realistically processed in a reasonable time and stored in one location. In such situations where time-bounded intensive computation analysis of shared large scale data is needed, one can try to use accessible computing and eventually storage resources distributed in different locations, according to the Grid model. Distributed computing and storage is not a new paradigm but until a few years ago networks were too slow, and the early days’ best-effort networking paradigm was not adequate to allow efficient use of remote resources. As the speed of networks increased and technical strategies for differentiation in service delivery became available, the interest in distributed computing has been taken to a new level. Thus, scaling the network and delivering bandwidth and services when and where a customer critical application needs it, are absolute prerequisites for success. A new network foundation is required, that will easily adapt to support the rapid growth, change and highly responsive service delivery required by Grid computing. In this scenario, the wide deployment of point-to-point wavelength division multiplexing (WDM) transmission systems in the Internet infrastructure has enhanced the bandwidth available in the network core by several orders of magnitude, introducing the need for faster switching in the optical domain. Consequently massive interest has focused on optical networking such that the answer to all the current performance open issues is conceived to lie in an intelligent dynamic photonic transport layer deployed in support of multiple, global, next generation Grid applications that are being designed and developed as highly asymmetric, highly distributed and resource intensive (e.g., computationally intensive, bandwidth intensive, etc.). New optical devices like WDM Multiplexers and Optical Cross-Connects (OXCs), and new control-plane protocols such as Generalized Multi-protocol Label Switching (GMPLS) [1] will make possible a pure photonic network where packets are routed through the network core without leaving the optical domain. The wavelength routing technology is considered extremely promising for the realization of new generation networks that will have to handle a significant portion of Grid applications’ traffic, demanding specific resources such as assured QoS. These applications must have powerful and flexible network integration capabilities for directly discovering and signalling for use of the networking resources that they require, triggering the dynamic provisioning of virtual traffic tunnels throughout the network, such as multi-wavelength lightpaths and multi-domain (electronic and optical) label/lambda switched paths (LSPs). In addition, there is a need to devise the proper mechanisms for QoS provisioning in wavelength routed networks that consider the physical characteristics and limitations of the optical domain without any loss in flexibility and performance. The best solution, at our advice, is applying optical technologies and electronic technologies in different spheres by implementing QoS at the edge of a network with electronic technologies and mapping the resulting QoS service classes into separate lightpaths in the network fully-optical core. In this paper we present a general framework for providing differentiated services QoS in photonic-based Grid infrastructures built on the strengths of GMPLS for dynamic path selection and wavelength assignment. State-of-the-art optical networking technology based on dynamic wavelength switching enables the creation of Grid services that allocate and release these paths either on-demand or by advance reservation. Specifically, we focus on the edge lambda switching routers (LSRs), located between label (electronic) switching domain and lambda (optical) switching domain. All the QoS requirements of the client Grid applications located on the network edge, and consequently committed in the label switching domain, will be fulfilled by the appropriate allocation of particular wavelengths on concatenated physical resources, or lightpaths, in the lambda switching domain. This framework makes it technically and economically viable to think of a set of computing and storage nodes connected through a high-performance optical network as one large computational device so that the boundaries between applications, computers, and networks truly dissolve. The paper is organized as follows. Section 2 introduces the technological scenario and the most important prerequisite concepts while Section 3 presents the main architectural details such as the transport network model and the Grid Service interface. Section 4 explains the QoS provisioning scheme that is analyzed through extensive simulation in Section 5, followed by the related works Section 6.