Showing papers on "Load balancing (computing) published in 2000"

Internet traffic engineering by optimizing OSPF weights

Abstract: Open shortest path first (OSPF) is the most commonly used intra-domain Internet routing protocol. Traffic flow is routed along shortest paths, splitting flow at nodes where several outgoing links are on shortest paths to the destination. The weights of the links, and thereby the shortest path routes, can be changed by the network operator. The weights could be set proportional to their physical distances, but often the main goal is to avoid congestion, i.e., overloading of links, and the standard heuristic recommended by Cisco is to make the weight of a link inversely proportional to its capacity. Our starting point was a proposed AT&T WorldNet backbone with demands projected from previous measurements. The desire was to optimize the weight setting based on the projected demands. We showed that optimizing the weight settings for a given set of demands is NP-hard, so we resorted to a local search heuristic. Surprisingly it turned out that for the proposed AT&T WorldNet backbone, we found weight settings that performed within a few percent from that of the optimal general routing where the flow for each demand is optimally distributed over all paths between source and destination. This contrasts the common belief that OSPF routing leads to congestion and it shows that for the network and demand matrix studied we cannot get a substantially better load balancing by switching to the proposed more flexible multi-protocol label switching (MPLS) technologies. Our techniques were also tested on synthetic internetworks, based on a model of Zegura et al., (1996), for which we did not always get quite as close to the optimal general routing.

On the impact of alternate path routing for load balancing in mobile ad hoc networks

Abstract: Alternate path routing (APR) can provide load balancing and route failure protection by distributing traffic among a set of diverse paths. These benefits make APR appear to be an ideal candidate for the bandwidth limited and mobile ad-hoc networks. However, we find that APR's potential is not fully realized in ad-hoc networks because of route coupling resulting from the geographic proximity of candidate paths between common endpoints. In multiple channel networks, coupling occurs when paths share common intermediate nodes. The coupling problem is much more serious in single channel networks, where coupling also occurs where one path crosses the radio coverage area of another path, The network's inherent route coupling is further aggravated by the routing protocol, which may provide an incomplete view of current network connectivity. Through analysis and simulation, we demonstrate the impact of route coupling on APR's delay performance in ad-hoc networks. In multiple channel environments, APR is able to provide a 20% reduction in end-to-end delay for bursty data streams. Though these gains are appreciable, they are about half what we would expect from APR with independently operating routes. Route coupling is so severe in single channel networks that APR provides only negligible improvements in quality of service.

World Wide Web caching: trends and techniques

Abstract: Academic and corporate communities have been dedicating considerable effort to World Wide Web caching. When correctly deployed, Web caching systems can lead to significant bandwidth savings, server load balancing, perceived network latency reduction, and higher content availability. We survey the state of the art in caching designs, presenting a taxonomy of architectures and describing a variety of specific trends and techniques.

Performance of hashing-based schemes for Internet load balancing

Abstract: Load balancing is a key technique for improving Internet performance. Effective use of load balancing requires good traffic distribution schemes. We study the performance of several hashing schemes for distributing traffic over multiple links while preserving the order of packets within a flow. Although hashing-based load balancing schemes have been proposed in the past, this is the first comprehensive study of their performance using real traffic traces. We evaluate five direct hashing methods and one table-based hashing method. We find that hashing using a 16-bit CRC over the five-tuple gives excellent load balancing performance. Further, load-adaptive table-based hashing using the exclusive OR of the source and destination IP addresses achieves comparable performance to the 16-bit CRC. Table-based hashing can also distribute traffic load according to unequal weights. We also report on four other schemes with poor to moderate performance.

A modular approach to packet classification: algorithms and results

Abstract: The ability to classify packets according to pre-defined rules is critical to providing many sophisticated value-added services, such as security, QoS, load balancing, traffic accounting, etc. Various approaches to packet classification have been studied in the literature with accompanying theoretical bounds. Practical studies with results applying to large number of filters (from 8K to 1 million) are rare. In this paper, we take a practical approach to the problem of packet classification. Specifically, we propose and study a novel approach to packet classification which combines a heuristic tree search with the use of filter buckets. Besides high performance and a reasonable storage requirement, our algorithm is unique in the sense that it can adapt to the input packet distribution by taking into account the relative filter usage. To evaluate our algorithms, we have developed realistic models of large scale filter tables, and used them to drive extensive experimentation. The results demonstrate the practicality of our algorithms for up to even 1 million filters.

Load-balancing clusters in wireless ad hoc networks

Abstract: Ad hoc networks consist of a set of identical nodes that move freely and independently and communicate with other node via wireless links. Such networks may be logically represented as a set of clusters by grouping together nodes that are in close proximity with one another. Clusterheads form a virtual backbone and may be used to route packets for nodes in their cluster. Nodes are assumed to have non-deterministic mobility pattern. Clusters are formed by diffusing node identities along the wireless links. Different heuristics employ different policies to elect clusterheads. Several of these policies are biased in favor of some nodes. As a result, these nodes shoulder greater responsibility and may deplete their energy faster, causing them to drop out of the network. Therefore, there is a need for load-balancing among clusterheads to allow all nodes the opportunity to serve as a clusterhead. We propose a load-balancing heuristic to extend the life of a clusterhead to the maximum budget before allowing the clusterhead to retire and give way to another node. This helps to evenly distribute the responsibility of acting as clusterheads among all nodes. Thus, the heuristic ensures fairness and stability. Simulation experiments demonstrate that the proposed heuristic does provide longer clusterhead durations than with no load-balancing.

Database load balancing for multi-tier computer systems

Abstract: A load balancing method and system for a transaction computer system having multiple database servers for at least one database, wherein database servers cooperate to provide a unified view of the data in the database. The method includes the steps of establishing connections to said multiple database servers for communicating with said database servers; and assigning transactions to respective ones of said multiple database servers to balance respective loads of said multiple database servers. Assigning each new transaction includes the steps of determining possible assignments of that new transaction to one or more of said multiple database servers, each said possible assignment to one of said multiple database servers being based on a load balancing scheme to balance respective loads of said multiple database servers; and assigning that new transaction to one of said multiple database servers as a function of said possible assignments of that new transaction, to balance respective loads of said multiple database servers.

How useful is old information

Abstract: We consider the problem of load balancing in dynamic distributed systems in cases where new incoming tasks can make use of old information. For example, consider a multiprocessor system where incoming tasks with exponentially distributed service requirements arrive as a Poisson process, the tasks must choose a processor for service, and a task knows when making this choice the processor queue lengths from T seconds ago. What is a good strategy for choosing a processor in order for tasks to minimize their expected time in the system? Such models can also be used to describe settings where there is a transfer delay between the time a task enters a system and the time it reaches a processor for service. Our models are based on considering the behavior of limiting systems where the number of processors goes to infinity. The limiting systems can be shown to accurately describe the behavior of sufficiently large systems and simulations demonstrate that they are reasonably accurate even for systems with a small number of processors. Our studies of specific models demonstrate the importance of using randomness to break symmetry in these systems and yield important rules of thumb for system design. The most significant result is that only small amounts of queue length information can be extremely useful in these settings; for example, having incoming tasks choose the least loaded of two randomly chosen processors is extremely effective over a large range of possible system parameters. In contrast, using global information can actually degrade performance unless used carefully; for example, unlike most settings where the load information is current, having tasks go to the apparently least loaded server can significantly hurt performance.

A comparison of load balancing techniques for scalable Web servers

Abstract: Scalable Web servers can be built using a network of workstations where server capacity can be extended by adding new workstations as the workload increases. The topic of our article is a comparison of different method to do load-balancing of HTTP traffic for scalable Web servers. We present a classification framework the different load-balancing methods and compare their performance. In addition, we evaluate in detail one class of methods using a prototype implementation with instruction-level analysis of processing overhead. The comparison is based on a trace driven simulation of traces from a large ISP (Internet Service Provider) in Norway. The simulation model is used to analyze different load-balancing schemes based on redirection of request in the network and redirection in the mapping between a canonical name (CNAME) and IP address. The latter is vulnerable to spatial and temporal locality, although for the set of traces used, the impact of locality is limited. The best performance is obtained with redirection in the network.

System for bypassing a server to achieve higher throughput between data network and data storage system

Abstract: A networked system is described in which the majority of data bypass the server(s). This design improves the end-to-end performance of network access by achieving higher throughput between the network and storage system, improving reliability of the system, yet retaining the security, flexibility, and services that a server-based system provides. The apparatus that provides this improvement consists of a network interface, server computer interface, and storage interface. It also has a switching element and a high-layer protocol decoding and control unit. Incoming traffic (either from the network or storage system) is decoded and compared against a routing table. If there is a matching entry, it will be routed, according to the information to the network, the storage interface, or sent to the server for further processing (default). The routing table entries are set up by the server based on the nature of the applications when an application or user request initially comes in. Subsequently, barring any changes or errors, there will be no data exchange between the server and the device (although, a control message may still flow between them). There may also be a speed matching function between the network and storage, load balancing function for servers, and flow control for priority and QoS purposes. Because the majority of data traffic will bypass the bus and the operating system (OS) of the server(s), the reliability and throughput can also be significantly improved. Therefore, for a given capacity of a server, much more data traffic can be handled. Certain improvements concerning one particular embodiment of the invention are also disclosed.

TeleMIP: telecommunications-enhanced mobile IP architecture for fast intradomain mobility

Abstract: This article first surveys existing protocols for supporting IP mobility and then proposes an extension to the mobile IP architecture, called TeleMIP. Our architecture attempts to achieve smaller handoff latency by localizing the scope of most location update messages within an administrative domain or a geographical region. TeleMIP is intended for use in evolving third-generation wireless networks, and introduces a new logical entity, called the mobility agent, which provides a mobile node with a stable point of attachment in a foreign network. While the MA is functionally similar to conventional foreign agents, it is located at a higher level in the network hierarchy than the subnet-specific FAs. Location updates for intradomain mobility are localized only up to the MA; transmission of global location updates are necessary only when the mobile changes administrative domains and/or geographical regions. By permitting the use of private or locally scoped addresses for handling intradomain mobility, TeleMIP allows efficient use of public address space. Also, by reducing the frequency of global update messages, our architecture overcomes several drawbacks of existing protocols, such as large latencies in location updates, higher likelihood of loss of binding update messages, and loss of inflight packets, and thus provides better mobility support for real-time services and applications. The dynamic creation of mobility agents (in TeleMLP) permits the use of load balancing schemes for the efficient management of network resources.

Linux Virtual Server for Scalable Network Services

Abstract: This paper describes the motivation, design, internal implementation of Linux Virtual Server. The goal of Linux Virtual Server is to provide a basic framework to build highly scalable and highly available network services using a large cluster of commodity servers. The TCP/IP stack of Linux kernel is extended to support three IP load balancing techniques, which can make parallel services of different kinds of server clusters to appear as a service on a single IP address. Scalability is achieved by transparently adding or removing a node in the cluster, and high availability is provided by detecting node or daemon failures and reconfiguring the system appropriately.

Load balancing method and system

Abstract: Load balancing method and system for balancing a processing load in a network, wherein load balancer upon receiving a client request selects one of a plurality of processing servers for serving the request. The selection of a processing server may be based on a location of a client and a processing server. The load balancer may comprise a load balancing master and a plurality of load balancing slaves, wherein the load balancing slaves receive client requests and transmit a selection request message to the load balancing master. The load balancing master selects a load balancing slave and/or a processing server for serving the request and generates a corresponding instruction message. The client request is served by the selected load balancing slave and the selected processing server by establishing a communication link between the client and the processing server. Service may involve execution of applications on the selected processing server under control of the client, for example word processors, scientific applications and similar.

Load balancing a cluster of web servers: using distributed packet rewriting

Abstract: We present and evaluate an implementation of a prototype scalable web server consisting of a load-balanced cluster of hosts that collectively accept and service TCP connections. The host IP addresses are advertised using round robin DNS (RR-DNS) technique, allowing any host to receive requests from any client. Once a client attempts to establish a TCP connection with one of the hosts, a decision is made as to whether or not the connection should be redirected to a different host-namely, the host with the lowest number of established connections. We use the low-overhead Distributed Packet Rewriting (DPR) technique to redirect TCP connections. In our prototype, each host keeps information about the remaining hosts in the system. Load information is maintained using periodic multicast amongst the cluster hosts. Performance measurements suggest that our prototype outperforms both pure RR-DNS and the stateless DPR solutions.

Optimizing static job scheduling in a network of heterogeneous computers

Abstract: This paper investigates static job scheduling schemes that distribute workload in a network of computers with different speeds. Static schemes involve very low overhead and complexity compared to dynamic schemes, but can still provide significant performance improvement over the case of no load balancing. Optimization techniques are proposed for workload allocation and job dispatching. Workload allocation is modeled as a non-linear optimization problem and solved mathematically. It is shown that allocating a disproportionately high percentage of jobs to the more powerful computers improves system performance. The proposed job dispatching algorithm is an extension of the traditional round-robin scheme. The objective is to reduce burstiness in the job arrival stream to each computer The schemes are evaluated by simulation experiments. Performance results verify their effectiveness in terms of mean response time, mean response ratio, and fairness. The Optimized Round-Robin (ORR) strategy which combines both techniques outperforms other static scheduling algorithms examined.

Apparatus and method for periodic load balancing in a multiple run queue system

Abstract: An apparatus and methods for periodic load balancing in a multiple run queue system are provided. The apparatus includes a controller, memory, initial load balancing device, idle load balancing device, periodic load balancing device, and starvation load balancing device. The apparatus performs initial load balancing, idle load balancing, periodic load balancing and starvation load balancing to ensure that the workloads for the processors of the system are optimally balanced.

Allocating Bandwidth for Bursty Connections

Abstract: In this paper, we undertake the first study of statistical multiplexing from the perspective of approximation algorithms. The basic issue underlying statistical multiplexing is the following: in high-speed networks, individual connections (i.e., communication sessions) are very bursty, with transmission rates that vary greatly over time. As such, the problem of packing multiple connections together on a link becomes more subtle than in the case when each connection is assumed to have a fixed demand. We consider one of the most commonly studied models in this domain: that of two communicating nodes connected by a set of parallel edges, where the rate of each connection between them is a random variable. We consider three related problems: (1) stochastic load balancing, (2) stochastic bin-packing, and (3) stochastic knapsack. In the first problem the number of links is given and we want to minimize the expected value of the maximum load. In the other two problems the link capacity and an allowed overflow probability p are given, and the objective is to assign connections to links, so that the probability that the load of a link exceeds the link capacity is at most $p$. In bin-packing we need to assign each connection to a link using as few links as possible. In the knapsack problem each connection has a value, and we have only one link. The problem is to accept as many connections as possible. For the stochastic load balancing problem we give an O(1)-approximation algorithm for arbitrary random variables. For the other two problems we have algorithms restricted to on-off sources (the most common special case studied in the statistical multiplexing literature), with a somewhat weaker range of performance guarantees. A standard approach that has emerged for dealing with probabilistic resource requirements is the notion of effective bandwidth---this is a means of associating a fixed demand with a bursty connection that "represents" its distribution as closely as possible. Our approximation algorithms make use of the standard definition of effective bandwidth and also a new one that we introduce; the performance guarantees are based on new results showing that a combination of these measures can be used to provide bounds on the optimal solution.

Interpreting stale load information

Abstract: In this paper, we examine the problem of balancing load in a large-scale distributed system when information about server loads may be stale. It is well-known that sending each request to the machine with the apparent lowest load can behave badly in such systems, yet this technique is common in practice. Other systems use round-robin or random selection algorithms that entirely ignore load information or that only use a small subset of the load information. Rather than risk extremely bad performance on one hand or ignore the chance to use load information to improve performance on the other, we develop strategies that interpret load information based on its age. Through simulation, we examine several simple algorithms that use such load interpretation strategies under a range of workloads. Our experiments suggest that by properly interpreting load information, systems can: 1) match the performance of the most aggressive algorithms when load information is fresh relative to the job arrival rate, 2) outperform the best of the other algorithms we examine by as much as 60 percent when information is moderately old, 3) significantly outperform random load distribution when information is older still, and 4) avoid pathological behavior even when information is extremely old.

System and method for load balancing

Abstract: A system for distributing load between multiple servers where more than one server in a sever cluster receives a request for connection from a client and each server makes a determination of whether or not to respond to the request Software modules running on the servers monitor and communicate relative abilities of each server to respond to client requests Each server responding to a percentage of client requests corresponding to its relative ability to respond

Fast and effective task scheduling in heterogeneous systems

Abstract: Recently (Proc. ACM Int. Conf. on Supercomput., 1999), we presented two very low-cost approaches to compile-time list scheduling where the tasks' priorities are computed statically or dynamically. For homogeneous systems, these two algorithms, called FCP (Fast Critical Path) and FLB (Fast Load Balancing), respectively, have been shown to yield a performance equivalent to other much more costly algorithms, such as MCP and ETF (Earliest Task First). In this paper, we present modified versions of FCP and FLB targeted at heterogeneous systems. We show that the modified versions yield a good overall performance, which is generally comparable to algorithms specifically designed for heterogeneous systems, such as HEFT (Heterogeneous Earliest Finish Time) or ERT (which are versions of MCP and ETF, respectively, using the task's completion time as the task priority). There are a few cases, mainly for irregular problems and large processor speed variance, where FCP's and FLB's performances drop to 32% and 63%, respectively. Considering the good overall performance and their very low cost, however, FCP and FLB are interesting options for scheduling very large problems on heterogeneous systems.

Parallel Multilevel Algorithms for Multi-Constraint Graph Partitioning

Abstract: Sequential multi-constraint graph partitioners have been developed to address the load balancing requirements of multi-phase simulations. The efficient execution of large multi-phase simulations on high performance parallel computers requires that the multi-constraint partitionings are computed in parallel. This paper presents a parallel formulation of a recently developed multi-constraint graph partitioning algorithm. We describe this algorithm and give experimental results conducted on a 128-processor Cray T3E. We show that our parallel algorithm is able to efficiently compute partitionings of similar edge-cuts as serial multi-constraint algorithms, and can scale to very large graphs. Our parallel multi-constraint graph partitioner is able to compute a three-constraint 128-way partitioning of a 7.5 million node graph in about 7 seconds on 128 processors of a Cray T3E.

Combining multilink and IP per-destination load balancing over a multilink bundle

Abstract: Load balanced transport of best efforts traffic together with delay-bounded traffic over a multilink bundle combines fragmentation and fragment distribution for best efforts packets with per-flow balancing for delay-bounded traffic. In the preferred embodiments, the best efforts packets receive Multilink Protocol treatment, including fragmentation and addition of sequence headers. Fragments of packets for one such communication go over the various links within the bundle, as appropriate for load balancing on the links. For each delay-bounded flow, such as for a VOIP service, the sending node hashes the packet header data and applies all packets for the flow to one of the links, assigned as a function of the hash value. Different flow headers produce different hash results; therefore the node sends different flows over the different links in a substantially balanced manner.

Method and apparatus to provide centralized call admission control and load balancing for a voice-over-IP network

Abstract: An admission control and load balancing system controls admission of packet streams or calls to a network and balances the packet traffic across the network, improving quality of service. The system includes a central database which stores information including cost data associated with individual paths and links across the network. A processor, in communication with the database, coordinates the admission control and load balancing decisions, and updates of the database cost data to reflect the dynamic network conditions, based on input from appropriate data sources. In one embodiment, referred to as the exact algorithm, the database is consulted by the admission control points or gatekeepers prior to admitting each arriving packet stream, and the database contents are updated call-by-call to reflect the allocation of resources to each admitted stream. In another embodiment, referred to as the inexact algorithm, control decision as well as database updates occur on a periodic rather than on a call-by-call basis to promote better scalability. In this embodiment, the processor periodically calculates admission decisions based on cost data in the central database. These admission decisions are then periodically forwarded to a satellite database associated with each gatekeeper, for storage and use in admission decisions until the next update epoch.

Geographic load balancing for scalable distributed Web systems

Abstract: Users of highly popular Web sites may experience long delays when accessing information. Upgrading content site infrastructure from a single node to a locally distributed Web cluster composed by multiple server nodes provides limited relief, because the cluster wide-area connectivity may become the bottleneck. A better solution is to distribute Web clusters over the Internet by placing content nodes in strategic locations. A geographically distributed architecture where the Domain Name System (DNS) servers evaluate network proximity and users are served from the closest cluster reduces network impact on response time. On the other hand, serving closest requests only may cause unbalanced servers and may increase system impact on response time. To achieve a scalable Web system, we propose to integrate DNS proximity scheduling with an HTTP request redirection mechanism that any Web server can activate. We demonstrate through simulation experiments that this further dispatching mechanism augments the percentage of requests with guaranteed response time, thereby enhancing the Quality of Service of geographically distributed Web sites. However, HTTP request redirection should be used selectively because the additional round-trip increases network impact on latency time experienced by users. As a further contribution, this paper proposes and compares various mechanisms to limit reassignments with no negative consequences on load balancing.

Highly scalable least connections load balancing

Abstract: A load balancing device according to an embodiment of the invention uses a predictor that comprises a plurality of Least Connections Control Blocks (LCCBs) that keeps track of the real servers with active connections. To speed up the search for the real server with the least number of active connections, an LCCB is kept for each metric. A metric is defined as the number of connections on a server divided by its weight (or capacity) of the server. This metric is kept as a quotient/remainder pair. The predictor sends out the real server address with the lowest metric whenever a new connection is required by the load balancing device.

Distribution Feeder Reconfiguration for Service Restoration and Load Balancing

Abstract: This paper describes two feeder reconfiguration algorithms for the purpose of service restoration and load balancing in a real-time operation environment. The developed methodologies combine optimization techniques with heuristic rules and fuzzy logic for efficiency and robust performance. Many of practical operating concerns of feeder reconfiguration and the coordination with other distribution automation applications are also addressed. The developed algorithms have been implemented as 8 production grade software. Test results on PG&E distribution feeders show that the performance is efficient and robust.

Wireless lan with load balancing

Abstract: A communication system with a plurality of access points (AP1, AP2, AP3) and at least one network station (5, 6), the network station (5, 6) being arranged to communicate with one of said plurality of access points (AP1, AP2, AP3) through a wireless communication protocol, each access point (AP1, AP2, AP3) is able to: monitor its access point traffic load and transmit an access point traffic load parameter (ATT) to the network station (5, 6), and the network station (5, 6) is able to: monitor its network station traffic load; store a network station traffic load parameter (AUTT); receive access point traffic load parameters (ATT) from the access points (AP1, AP2, AP3); select a communication connection with one of the access points (AP1, AP2, AP3) using a predetermined cost function taking the access point traffic load parameters (ATT) and the network station traffic load parameters (AUTT) into account.

A New Paradigm for Parallel Adaptive Meshing Algorithms

Abstract: We present a new approach to the use of parallel computers with adaptive finite element methods. This approach addresses the load balancing problem in a new way, requiring far less communication than current approaches. It also allows existing sequential adaptive PDE codes such as PLTMG and MC to run in a parallel environment without a large investment in recoding. In this new approach, the load balancing problem is reduced to the numerical solution of a small elliptic problem on a single processor, using a sequential adaptive solver, without requiring any modifications to the sequential solver. The small elliptic problem is used to produce a posteriori error estimates to predict future element densities in the mesh, which are then used in a weighted recursive spectral bisection of the initial mesh. The bulk of the calculation then takes place independently on each processor, with no communication, using possibly the same sequential adaptive solver. Each processor adapts its region of the mesh independently, and a nearly load-balanced mesh distribution is usually obtained as a result of the initial weighted spectral bisection. Only the initial fan-out of the mesh decomposition to the processors requires communication. Two additional steps requiring boundary exchange communication may be employed after the individual processors reach an adapted solution, namely, the construction of a global conforming mesh from the independent subproblems, followed by a final smoothing phase using the subdomain solutions as an initial guess. We present a series of convincing numerical experiments which illustrate the effectiveness of this approach. The justification of the initial refinement prediction step, as well as the justification of skipping the two communication-intensive steps, is supported by some recent [J. Xu and A. Zhou, Math. Comp., to appear] and not so recent [J. A. Nitsche and A. H. Schatz, Math. Comp., 28 (1974), pp. 937--958; A. H. Schatz and L. B. Wahlbin, Math. Comp., 31 (1977), pp. 414--442; A. H. Schatz and L. B. Wahlbin, Math. Comp., 64 (1995), pp. 907--928] results on local a priori and a posteriori error estimation.

A Unified Algorithm for Load-balancing Adaptive Scientific Simulations

Abstract: Adaptive scientific simulations require that periodic repartitioning occur dynamically throughout the course of the computation. The repartitionings should be computed so as to minimize both the inter-processor communications incurred during the iterative mesh-based computation and the data redistribution costs required to balance the load. Recently developed schemes for computing repartitionings provide the user with only a limited control of the tradeoffs among these objectives. This paper describes a new Unified Repartitioning Algorithm that can tradeoff one objective for the other dependent upon a user-defined parameter describing the relative costs of these objectives. We show that the Unified Repartitioning Algorithm is able to reduce the precise overheads associated with repartitioning as well as or better than other repartitioning schemes for a variety of problems, regardless of the relative costs of performing inter-processor communication and data redistribution. Our experimental results show that this scheme is extremely fast and scalable to large problems.