Popular Web sites cannot rely on a single powerful server nor on independent mirrored-servers to support the ever-increasing request load. Distributed Web server architectures that transparently schedule client requests offer a way to meet dynamic scalability and availability requirements. The authors review the state of the art in load balancing techniques on distributed Web-server systems, and analyze the efficiencies and limitations of the various approaches.

Dynamic load balancing on Web-server systems

In this paper we address power conservation for clusters of workstations or PCs. Our approach is to develop systems that dynamically turn cluster nodes on – to be able to handle the load imposed on the system efficiently – and off – to save power under lighter load. The key component of our systems is an algorithm that makes load balancing and unbalancing decisions by considering both the total load imposed on the cluster and the power and performance implications of turning nodes off. The algorithm is implemented in two different ways: (1) at the application level for a cluster-based, localityconscious network server; and (2) at the operating system level for an operating system for clustered cycle servers. Our experimental results are very favorable, showing that our systems conserve both power and energy in comparison to traditional systems.

/pdf/load-balancing-and-unbalancing-for-power-and-performance-in-2bxenxgtur.pdf

Load balancing and unbalancing for power and performance in cluster-based systems

So far virtual machine (VM) migration has focused on transferring the run-time memory state of the VMs in local area networks (LAN). However, for wide-area network (WAN) migration it is crucial to not just transfer the VMs image but also transfer its local persistent state (its file system) and its on-going network connections. In this paper we address both: by combining a block-level solution with pre-copying and write throttling we show that we can transfer an entire running web server, including its local persistent state, with minimal disruption --- three seconds in the LAN and 68 seconds in the WAN); by combining dynDNS with tunneling, existing connections can continue transparently while new ones are redirected to the new network location. Thus we show experimentally that by combining well-known techniques in a novel manner we can provide system support for migrating virtual execution environments in the wide area.

Live wide-area migration of virtual machines including local persistent state

The previous research on cluster-based servers has focused on homogeneous systems. However, real-life clusters are almost invariably heterogeneous in terms of the performance, capacity, and power consumption of their hardware components. In this paper, we argue that designing efficient servers for heterogeneous clusters requires defining an efficiency metric, modeling the different types of nodes with respect to the metric, and searching for request distributions that optimize the metric. To concretely illustrate this process, we design a cooperative Web server for a heterogeneous cluster that uses modeling and optimization to minimize the energy consumed per request. Our experimental results for a cluster comprised of traditional and blade nodes show that our server can consume 42% less energy than an energy-oblivious server, with only a negligible loss in throughput. The results also show that our server conserves 45% more energy than an energy-conscious server that was previously proposed for homogeneous clusters.

Energy conservation in heterogeneous server clusters

In this chapter we address power conservation for clusters of workstations or PCs. Our approach is to develop systems that dynamically turn cluster nodes on - to be able to handle the load imposed on the system efficiently - and off - to save power under lighter load. The key component of our systems is an algorithm that makes cluster reconfiguration decisions by considering the total load imposed on the system and the power and performance implications of changing the current configuration. The algorithm is implemented in two common cluster-based systems: a network server and an operating system for clustered cycle servers. Our experimental results are very favorable, showing that our systems conserve both power and energy in comparison to traditional systems.

Dynamic cluster reconfiguration for power and performance

To construct high performance Web servers, system builders are increasingly turning to distributed designs. An important challenge that arises in such designs is the need to direct incoming connections to individual hosts. Previous methods for connection routing (layer 4 switching) have employed a centralized node to handle all incoming requests. In contrast, we propose a distributed approach, called distributed packet rewriting (DPR), in which all hosts of the distributed system participate in connection routing. DPR promises better scalability and fault-tolerance than the current practice of using centralized special-purpose connection routers. We describe the implementation of four variants of DPR and compare their performance. We show that DPR provides performance comparable to centralized alternatives, measured in terms of throughput and delay. Also, we show that DPR enhances the scalability of Web server clusters by eliminating the performance bottleneck exhibited when centralized connection routing techniques are utilized.

/pdf/distributed-packet-rewriting-and-its-application-to-scalable-14zse6p4cp.pdf

D. Martin

Papers

Distributed packet rewriting and its application to scalable server architectures