This paper presents an Erlang reduced load model to analyze Optical Burst Switched Grid networks. The model allows the evaluation of job blocking probabilities, in which blocking occurs in both the transport network and the resources where jobs are processed. Additionally, a novel routing strategy is introduced to improve the usage efficiency of the existing infrastructure. Simulation analysis is used to confirm the validity and accuracy of the model, and the effectiveness of the novel routing strategy is shown for Grids over OBS.

When deploying Grid infrastructure, the problem of dimensioning arises: how many servers to provide, where to place them, and which network to install for interconnecting server sites and users generating Grid jobs? In contrast to classical optical network design problems, it is typical of optical Grids that the destination of traffic (jobs) is not known beforehand. This leads to so-called anycast routing of jobs. For network dimensioning, this implies the absence of a clearly defined (source,destination)-based traffic matrix, since only the origin of Grid jobs (and their data) is known, but not their destination. The latter depends not only on the state of Grid resources, including network, storage, and computational resources, but also the Grid scheduling algorithm used. We present a phased solution approach to dimension all these resources, and use it to evaluate various scheduling algorithms in two European network case studies. Results show that the Grid scheduling algorithm has a substantial impact on the required network capacity. This capacity can be minimized by appropriately choosing a (reasonably small) number of server site locations: an optimal balance can be found, in between the single server site case requiring a lot of network traffic to this single location, and an overly fragmented distribution of server capacity over too many sites without much statistical multiplexing opportunities, and hence a relatively large probability of not finding free servers at nearby sites.

Abstract. In light of recently stirred energy consumption concerns, we investigate the opportunities for power consumption reduction in Grids. Considering real life Grid traces, we note considerable fluctuations in load. We consider a peak load dimensioning strategy to derive how much servers to install in computational Grids. In lower loaded periods, there is a potential to save energy by dynamically powering on/off servers to address the actual demand for computational capacity. An appropriate scheduling and power-saving scheme can, under such lower-load conditions, considerably reduce energy consumption. The price paid is that some jobs are executed at sites more remote than closer powered-down ones. Yet, the resulting penalty in consumed bandwidth is rather limited and is expected not to cancel the power consumption advantage. Key words: Grids, Green ICT, Power-awareness, Grid scheduling. 1

Abstract-When deploying Grid infrastructure, the problem of dimensioning arises: how many servers to provide, where to place them, and which network to install for interconnecting server sites and users generating Grid jobs? In contrast to classical optical network design problems, it is typical of optical Grids that the destination of traffic (jobs) is not known beforehand. This leads to so-called anycast routing of jobs. For network dimensioning, this implies the absence of a clearly defined (source,destination)-based traffic matrix, since only the origin of Grid jobs (and their data) is known, but not their destination. The latter depends not only on the state of Grid resources, including network, storage, and computational resources, but also the Grid scheduling algorithm used. We present a phased solution approach to dimension all these resources, and use it to evaluate various scheduling algorithms in two European network case studies. Results show that the Grid scheduling algorithm has a substantial impact on the required network capacity. This capacity can be minimized by appropriately choosing a (reasonably small) number of server site locations: an optimal balance can be found, in between the single server site case requiring a lot of network traffic to this single location, and an overly fragmented distribution of server capacity over too many sites without much statistical multiplexing opportunities, and hence a relatively large probability of not finding free servers at nearby sites.

Abstract—Notwithstanding IP anycast’s introduction in Internet standards dates back to 1993 and its more recent adoption in IPv6 standards, its use in production environments is limited to date. This is mainly because native IP anycast lacks routing scalability and does not support session-based communications, thereby limiting its applicability to single request-response services such as DNS. For this reason, we propose a transparent anycast overlay architecture that retains the strengths of native anycast and neutralizes above-mentioned limitations. The resulting proxy infrastructure unleashes the power of anycast by opening up new opportunities for transparent distributed service provisioning. Taking into account user demands, available resources, network overhead and anycast infrastructure costs, we provide nearoptimal heuristics for the placement of proxy nodes and dimensioning the infrastructure in large networks. We show that even modest overlay infrastructures, consisting of a small number of proxy routers, provide an effective stateful anycast solution where the detour via the proxy routers is negligible in terms of extra network load. Furthermore, simulation results illustrate that server state aggregation in the proxy nodes lessens control plane overhead, which contributes significantly to service robustness. I.

Start date of project: 01/09/2005 Duration: 36 months

An important problem in optical networking is to dimension the network: given the amount of traffic to carry, determine the required amount of network resources (esp. wavelengths). In traditional scenarios, the traffic is specified in terms of a (source,destination)-based traffic matrix. In an optical Grid scenario however, the anycast principle applies: users submit jobs, and generally do not care where exactly they end up being executed. Thus, the destination of traffic is not known beforehand and traditional dimensioning algorithms are not directly applicable. On the other hand, this flexibility in choosing a destination opens opportunities to save on backup network resources: to protect against failures, we can opt to redirect jobs to another location (i.e. exploit relocation). In this paper we (i) outline how to derive a traffic matrix in a step-wise grid dimensioning approach, and (ii) present an assessment of potential network resource savings in resilient network dimensioning by exploiting relocation.

ABSTRACT In optical Grid networks, the main challenge is to account for not only network parameters, but also for resource availability. Anycast routing has previously been proposed as an effective solution to provide job scheduling services in optical Grids, offering a generic interface to access Grid resources and services. The main weakness of this approach is its limited scalability, especially in a multi-domain scenario. This paper proposes a novel anycast proxy architecture, which extends the anycast principle to a multi-domain scenario. The main purpose of the architecture is to perform aggregation of resource and network states, and as such improve computational scalability and reduce control plane traffic. Furthermore, the architecture has the desirable properties of allowing Grid domains to maintain their autonomy and hide internal configuration details from other domains. Finally, we propose an impairment-aware anycast routing algorithm that incorporates the main physical layer characteristics of large-scale optical networks into its path computation process. By integrating the proposed routing scheme into the introduced architecture we demonstrate significant network performance improvements. Keywords: Grid computing, routing algorithms, optical networks, physical impairments, anycast routing. INTRODUCTION Today, the need for network systems to support storage and computing services for science and business, is often satisfied by relatively isolated computing infrastructure (clusters). Migration to truly distributed and integrated applications requires optimization and (re)design of the underlying network technology to create a Grid platform for the cost and resource efficient delivery of network services with substantial data transfer, processing power and/or data storage requirements. Optical networks offer an undeniable potential for the Grid, given their proven track-record in the context of high-speed, long-haul, networking. Not only eScience applications dealing with large experimental data sets (e.g. particle physics) but also business/consumer oriented applications can benefit from optical Grid infrastructure [1]: both the high data rates typical of eScience applications and the low latency requirements of consumer/business applications (cf. interactivity) can effectively be addressed. When using transparent WDM as such network technology, signals are transported end-to-end optically without being converted to the electrical domain in between. Connection provisioning of all-optical connections (lightpaths) between source and destination nodes is based on specific routing and wavelength assignment algorithms (RWA). Traditional RWA schemes only account for network conditions such as connectivity and available capacity, without considering physical layer details. However, in transparent optical networks covering large geographical areas, the optical signal experiences the accumulation of physical impairments through transmission and switching, possibly resulting in unacceptable signal quality Another emerging and challenging task in distributed and heterogeneous computing environments, is job scheduling: when and where to execute a given Grid job, based on the requirements of the job (for instance a deadline and minimal computational power) and the current state of the network and resources. Traditionally, a local scheduler optimizes utilization and performance of a single Grid site, while a meta-scheduler is distributes workload across different sites. Current implementations of these (meta-)schedulers only account for Grid resource availability In this paper we propose a novel architecture to support impairment-aware anycast routing for large-scale optical Grid networks. Section 2 discusses general approaches to support multi-domain networks. We then proceed to introduce a novel architecture, which can provide anycast Grid services in a multi-domain scenario (Section 3). Simulation analysis is used to demonstrate the improved scalability without incurring significant performance loss. Furthermore, Section 4 shows how to incorporate physical layer impairments, to further improve the performance of optical Grid networks. Conclusions are presented in Section 5.