Mike H. MacGregor

Load balancing for parallel forwarding

Workload distribution is critical to the performance of network processor based parallel forwarding systems. Scheduling schemes that operate at the packet level, e.g., round-robin, cannot preserve packet-ordering within individual TCP connections. Moreover, these schemes create duplicate information in processor caches and therefore are inefficient in resource utilization. Hashing operates at the flow level and is naturally able to maintain per-connection packet ordering; besides, it does not pollute caches. A pure hash-based system, however, cannot balance processor load in the face of highly skewed flow-size distributions in the Internet; usually, adaptive methods are needed. In this paper, based on measurements of Internet traffic, we examine the sources of load imbalance in hash-based scheduling schemes. We prove that under certain Zipf-like flow-size distributions, hashing alone is not able to balance workload. We introduce a new metric to quantify the effects of adaptive load balancing on overall forwarding performance. To achieve both load balancing and efficient system resource utilization, we propose a scheduling scheme that classifies Internet flows into two categories: the aggressive and the normal, and applies different scheduling policies to the two classes of flows. Compared with most state-of-the-art parallel forwarding schemes, our work exploits flow-level Internet traffic characteristics.

Development and performance assessment of a distributed asynchronous protocol for real-time network restoration

The methodology used, and results obtained, in the development and verification of a protocol for real-time network restoration are described. This protocol, called selfhealing, relies on a combination of hardware and software features to achieve an advance in the speed of restoring network spans that have been cut. The hardware environment provides a paradigm for heavily parallel, asynchronous, distributed interaction. This environment is uniquely suited to interactions between digital cross-connect switches (DCS) embedded in a highly-capacity transport network. A speed determination based on an implementation of the protocol exactly as it would exist in the target DCS host machine is given. >

Deficits for bursty latency-critical flows: DRR++

Fair queuing was invented to ensure that every flow gets its fair share of the total bandwidth. Efficient fair queuing using deficit round-robin, DRR, proposed by Shreedhar and Varghese (see IEEE/ACM Trans. Net., vol.4, no.4, p.386-97, 1996), reduces the work to process each packet from O(log(n)) to O(1). DRR+ was also extended to accommodate latency-critical flows. DRR+ uses a timer to police each latency-critical flow and was shown to have a latency bound of (n/sub c/s)+(M/B) for these flows. The definition of the contract of Shreedhar and Varghese, however, constrains a latency-critical flow to generate very smooth arrivals. By giving another definition of contract, we return to using the original concept of deficit to enforce each flows commitment to its contract. This allows for bursty arrivals which may occur either as the result of source bursts, or as a result of the dynamics of multihop network paths.

Optimal Spare Capacity Preconfiguration for FasterRestoration of Mesh Networks

Several distributed real-time methods have beenproposed for restoration from single span failures indigital transport networks. These methods have thepotential to avoid user service outages due to such failures, if they operate quickly enough. Forexample, switched 64 kbps connections will not bedisconnected if the network can be restored before thetime at which calls in progress are dropped, typically 1-2 seconds after a failure. However, it willbe difficult to achieve the goal of sub-secondrestoration if cross-connects cannot operate crosspointsquickly enough, either due to large workloads during a restoration response, or because ofimplementation choices such as testing eachcross-connection while in the midst of a serious outage.The results in this paper demonstrate that it can beuseful to pre-operate selected cross-points between thespare links of a mesh-restorable network before anyfailure has occurred, putting the network into astatistically optimal state of readiness. When a failure occurs, some of the preconfigured restorationpath bundles can be used immediately. If morerestoration paths are needed, they can be obtained by areal-time restoration process. The first advantage of preconfiguration is that the number ofcross-connection operations may be greatly reduced oreliminated for a portion of the affected traffic. Thiswill reduce restoration time significantly. Secondly, after utilizing preconfigured restorationpaths, the workload of a real-time restoration processwill be lower because it will be searching for fewerpaths. This paper demonstrates that preconfiguration can supply a significant proportion of thereplacement capacity required after a span failure. Theresults are obtained through integerprogramming.

An efficient scheme to remove crawler traffic from the Internet

We estimate that approximately 40% of current Internet traffic is due to Web crawlers retrieving pages for indexing. We address this problem by introducing an efficient indexing system based on active networks. Our approach employs strategically placed active routers that constantly monitor passing Internet traffic, analyze it, and then transmit the index data to a dedicated back-end repository. Our simulations have shown that active indexing is up to 30% more efficient than the current crawler-based techniques.

A scalable load balancer for forwarding internet traffic: exploiting flow-level burstiness

Packet scheduling in parallel forwarding systems is a hard problem. Two major goals of a scheduler that distributes incoming packets to multiple forwarding engines are to achieve high system utilization (by balancing the load evenly among the multiple engines) and to maintain packet ordering within individual flows. Additionally, from the viewpoint of the overall performance, the system should exhibit a good cache behavior by preserving temporal locality in the workload of each forwarding engine. In this paper, we show how the burstiness in Internet flows can be exploited to improve the performance of the scheduler. Specifically, TCP flows, which contribute to over 90 percent of the Internet traffic, transmit in bursts with relatively large delays in between. We propose a load balancing scheme based on this insight to achieve the scheduling goals. Our design is verified by simulations driven by real-world traces.

Mesh/arc networking: an architecture for efficient survivable self-healing networks

This paper summarises the results of a study which compares two architectures for survivable self-healing transport networks. Mesh/ring and mesh/arc networks are compared in terms of equipment counts required to achieve fully survivability against single span failure. Heuristic design procedures are developed. The mesh/arc network architecture is shown to be superior in all cases studied. Two main contributions to survivable network design/planning have resulted from this study. The first is the use of the network nodal degree distribution to partition a network into core and access regions. The second is the use of span elimination techniques as a cost-reducing transformation for the access region of networks.<<ETX>>

Distributed optimal dynamic base station positioning in wireless sensor networks

Base station (BS) positioning is an effective method for improving the performance of wireless sensor networks (WSNs). A metric-aware optimal BS positioning and relocation mechanism for WSNs is proposed. This technique locates the BS with respect to the available resources and the amount of traffic traveling through the sensor nodes. The BS calculates its own position over time in response to the dynamic environment in which the sensor nodes operate. In most WSN environments, communication channel experiences nonlinearity that is influenced by path loss, attenuation of signal as it propagated through space, greater than 2. In this work, we solved the problem of BS positioning in nonlinear environments. We propose a weighted linear or nonlinear least squares optimization depending on the value of the path loss exponent. We also propose a distributed algorithm that can effectively handle the required computation by exploiting node cooperation. The goal is to minimize the total energy consumption and to prolong lifetime of the WSNs. The performance of the proposed technique is evaluated for various network setups and conditions. Our simulation results demonstrate that BS positioning and relocation can significantly improve the lifetime and power efficiency in WSNs.

Combining Petri nets and ns-2: a hybrid method for analysis and simulation

Network protocol performance and evaluation study is often carried out using a couple of widely used simulation toolkits, such as ns-2 and OPNET. These toolkits usually contain many built-in protocol models. Using these toolkits is very efficient due to the abundant models. However, the correctness of a protocol that interests us can never be proved by simulation itself. Petri net modeling enables us to verify the protocol of interest formally. But because of the generality of Petri nets, not many network protocol models are bundled with Petri net modeling tools. In this paper we present an innovative network simulation methodology that benefits from the interaction between ns-2 and Petri nets. A communication mechanism based on Socket Programming and a synchronization mechanism used to coordinate ns-2 and Petri nets were designed to make possible the interaction. In this paper, a new version of SACK TCP, alpha-min Paced SACK TCP, is used to illustrate the power of the proposed methodology

An Adaptive Load Balancer for Multiprocessor Routers

By investigating flow-level characteristics of Internet traffic, the authors are able to trace the root of load imbalance in hash-based load-splitting schemes. They model flow popularity distributions as Zipf-like and prove that for typical Internet traffic, a hashing scheme cannot balance workload statistically, not even in the long run. They then develop a novel load-balancing packet scheduler for parallel forwarding systems. The scheduler capitalizes on the nonuniform flow reference pattern and especially the presence of a few high-rate flows in Internet traffic. The authors show that detecting and scheduling these flows can be very effective in balancing workloads among network processors. They introduce an important metric, adaptation disruption, to measure the scheduling efficiency of load-balancing mechanisms in parallel forwarding systems. Because there are relatively few large flows, reassigning them in the load balancer results in little disruption to the states of individual processors. The ideas are validated by simulation results. Finally, the authors discuss the effects on cache performance when classifying flows using two different flow definitions: the destination IP address and the five-tuple. The latter results in finer flow granularity but worse route cache hit rate, which can lead to the degradation of routing table lookup performance.