Anshul Kantawala

Link buffer sizing: a new look at the old problem

We revisit the question of how much buffer an IP router should allocate for its droptail FIFO link. For a long time, setting the buffer size to the bitrate-delay product has been regarded as reasonable. Recent studies of interaction between queueing at IP routers and TCP congestion control offered alternative guidelines. First, we explore and reconcile contradictions between the existing rules. Then, we argue that the problem of link buffer sizing needs a new formulation: design a buffer sizing algorithm that accommodates needs of all Internet applications without engaging IP routers in any additional signaling. Our solution keeps network queues short: set the buffer size to 2L datagrams, where L is the number of input links. We also explain how end systems can utilize the network effectively despite such small buffering at routers.

An architecture for monitoring, visualization, and control of gigabit networks

We outline a design of a highly scalable network monitoring, visualization, and control (NMVC) system with advanced algorithmic and human-in-the-loop capability. This capability allows network administrators to calibrate and fine-tune network and application parameters in real time according to observed traffic patterns. The goal of the NMVC system is to ensure adequate quality of service to network users, while maintaining high network resource utilization. The main components of our system are: a network probe and an endsystem probe which can probe gigabit/s links, software network management agents that provide extensible multi-attribute event filtering for highly scalable date/event collection, efficient online event ordering algorithms that can help synthesize and display a consistent view of network health, status, and performance and a View Choreographer that allows management applications and administrators to specify the mapping of network events to higher-level events and to visualization objects and updates.

Design and evaluation of a high-performance dynamically extensible router

This paper describes the design, implementation and performance of an open, high performance, dynamically extensible router under development at Washington University in St. Louis. This router supports the dynamic installation of software and hardware plug-ins in the data path of application data flows. It provides an experimental platform for research on programmable networks, protocols, router software and hardware design, network management, quality of service and advanced applications. It is designed to be flexible without sacrificing performance. It supports gigabit links and uses a scalable architecture suitable for supporting hundreds or even thousands of links. The systems flexibility makes it an ideal platform for experimental research on dynamically extensible networks that implement higher level functions in direct support of individual application sessions.

Queue management for short-lived TCP flows in backbone routers

Packets in the Internet can experience large queueing delays during busy periods. Backbone routers are generally engineered to have large buffers, in which packets may wait as long as half a second (assuming FIFO service, longer otherwise). During congestion periods, these buffers may stay close to full, subjecting packets to long delays, even when the intrinsic latency of the path is relatively small. This paper studies the performance improvements that can be obtained for short-lived TCP flows by using more sophisticated packet schedulers, than are typical of Internet routers. The results show that the large buffers found in WAN routers contribute only marginally to improving router throughput, and the higher delays that come with large buffers makes them a dubious investment. The results also show that better packet scheduling algorithms can produce dramatic improvements in fairness. Using ns-2 simulations, we show that algorithms using multiple queues can significantly outperform RED and Blue, especially at smaller buffer sizes. Given a traffic mix of short-lived TCP flows with different round-trip times, longer round-trip time flows achieve 80% of their fair-share using multiqueue schedulers, compared to 40% under RED and Blue. We observe a similar performance improvement for multi-hop paths. We also show that performance results can be reliably scaled across a wide range of parameter values, so long as the ratio of the buffer size to the link bandwidth-delay product is held invariant.

Simulation Perspectives on Link Buffer Sizing

The problem of determining the optimal buffer size for an Internet link has recently attracted the strong revived attention of networking scientists. While consonant in dissenting from the conventional wisdom to set the buffer size to the product of the link bitrate and round-trip propagation delay of served connections, the new studies often propose alternative guidelines that starkly contradict each other. In this paper, we review the problem of link buffer sizing from two simulation perspectives. First, we use the packet-level simulator ns-2, which is currently the predominant tool for evaluation of network designs, to explore the contradictions between the buffer sizing guidelines. We attribute the contradictions to differences in assumptions and goals of the earlier studies and present our own argument for using small constant buffer sizes. Second, we reflect on the suitability of ns-2 itself for studying the problem of link buffer sizing. We the identify ns-2 deficiencies that undermine the trustworthiness of results provided by this prevailing simulation tool. We also suggest directions and discuss challenges for designing a simulation methodology capable of providing undisputed answers on link buffer sizing.

Supporting DIS applications using ATM multipoint connection caching

This report describes an ATM multipoint connection caching strategy (AMCC) to control the explosive growth of traffic within the network and at an endpoint in a large distributed interactive simulation (DIS) application such as a battlefield simulation. For very large DIS applications with 100000 entities, the current method of broadcasting information among entities will no longer be feasible due to the computational and network bandwidth limitations. Our scheme divides the simulation space into grids and each grid square or a set of grid squares forms a multicast group. Entities join the groups within their perception range and thus, they receive state updates only from entities within their perception range. The AMCC utilizes the unique capabilities of our dynamic multipoint ATM signaling protocol, CMAP. We have implemented the AMCC and have successfully demonstrated a small DIS battlefield simulation application running over our ATM testbed.

Efficient Queue Management for TCP Flows

Packets in the Internet can experience large queueing delays during busy periods. Backbone routers are generally engineered to have large buffers, in which packets may wait as long as half a second (assuming FIFO service, longer otherwise). During congestion periods, these buffers may stay close to full, subjecting packets to long delays, even when the intrinsic latency of the path is relatively small. This paper studies the performance improvements that can be obtained by using more sophisticated packet schedulers, than are typical of Internet routers. The results show that the large buffers found in WAN routers contribute only marginally to improving router throughput, and the higher delays that come with large buffers makes them a dubious investment. The results also show that better packet scheduling algorithms can produce dramatic improvements in fairness. Using ns-2 simulations, we show that algorithms using multiple queues can significantly outperform RED and Blue, especially at smaller buffer sizes. Over a single-bottleneck link, the variance in TCP goodput using the proposed multiqueue packet schedulers is one-tenth that obtained with RED and one-fifth that obtained with Blue. Given a traffic mix of TCP flows with different round-trip times, longer round-trip time flows achieve of their fair-share using multiqueue schedulers, compared to under RED and Blue. We observe a similar performance improvement for multi-hop paths.

Intelligent Packet Discard Policies for Improved TCP Queue Management

Recentstudieshave shown thatsuitably-designedpacket discardpoliciescandramatically improve theperformanceof fair queueingmechanismsin internetrouters.TheQueueState Deficit RoundRobinalgorithm(QSDRR)preferentiallydiscardsfrom long queues, but introduceshysteresisinto thediscardpolicy to minimizesynchronizationamongTCPflows. QSDRRprovideshigherthroughputandmuchbetterfairnessthansimplerqueueingmechanisms,suchasTail-Drop,REDandBlue. However, becauseQSDRRdiscardspacketsthat have previously beenqueued,it cansignficantlyincreasethememorybandwidthrequirementsof high performancerouters. In this paper , we explorealternati vesto QSDRRthat provide comparableperformance, while allowing packetsto bediscardedonarrival, saving memorybandwidth.Usingns-2simulations,weshow thattherevisedalgorithmscancome closeto matchingtheperformanceof QSDRRandsubstantiallyoutperformREDandBlue. Given a traffic mix of TCP flows with different round-trip times, longer round-trip time flows achieve of their fair-shareusingtherevisedalgorithms,comparedto under RED andBlue. We observe a similar improvementin fairnessfor long multi-hop paths competingagainstshortcross-traf fic paths.Wealsoshow thatthesealgorithmscanprovide goodperformance, wheneachqueueis sharedamongmultiple flows. This work is supportedin partby DARPA GrantN660001-01-1-8930 Intelligent Packet Discard Policiesfor Impr ovedTCP QueueManagement AnshulKantawalaandJonathanTurner Departmentof ComputerScienceandEngineering WashingtonUniversity St. Louis,MO 63130 anshul,jst @arl.wustl.edu

Improving the performance of internet data transport

With the explosion of the World Wide Web, the Internet infrastructure faces new challenges in providing high performance for data traffic. First, it must be able to provide a fair-share of congested link bandwidth to every flow. Second, since Web traffic is inherently interactive, it must minimize the delay for data transfer. Recent studies have shown that queue management algorithms such as Tail Drop, RED and Blue are deficient in providing high throughput, low delay paths for a data flow. Two major shortcomings of the current algorithms are: they allow TCP flows to get synchronized and thus require large buffers during congestion to enable high throughput; and they allow unfair bandwidth usage for shorter round-trip time TCP flows. We propose algorithms using multiple queues and discard policies with hysteresis at bottleneck routers to address both these issues. Using ns-2 simulations, we show that these algorithms can significantly outperform RED and Blue, especially at smaller buffer sizes. Using multiple queues raises two new concerns: scalability and excess memory bandwidth usage caused by dropping packets which have been queued. We propose and evaluate an architecture using Bloom filters to evenly distribute flows among queues to improve scalability. We have also developed new intelligent packet discard algorithms that discard packets on arrival and are able to achieve performance close to that of policies that may discard packets that have already been queued. Finally, we propose better methods for evaluating the performance of fair-queueing methods. In the current literature, fair-queueing methods are evaluated based on their worst-case performance. This can exaggerate the differences among algorithms, since the worst-case behavior is dependent on the precise timing of packet arrivals. This work seeks to understand what happens under more typical circumstances.

Selecting the Buffer Size for an IP Network Link

In this paper, we revisit the problem of selecting the buffer size for an IP network link. After a comprehensive overview of issues relevant to the link buffer sizing, we examine usefulness of existing guidelines for choosing the buffer size. Our analysis shows that the existing recommendations not only are difficult to implement in the context of IP networks but also can severely hurt interactive distributed applications. Then, we argue that the networking research community should change its way of thinking about the link buffer sizing problem: the focus should shift from optimizing performance for applications of a particular type to maximizing diversity of application types that IP networks can support effectively. To achieve this new objective, we propose using small buffers for IP network links. Selecting the Buffer Size for an IP Network Link Sergey Gorinsky and Anshul Kantawala and Jonathan Turner Applied Research Laboratory Department of Computer Science and Engineering Washington University in St. Louis St. Louis, MO 63130-4899, USA gorinsky,anshul,jst @arl.wustl.edu