Showing papers presented at "International Workshop on Quality of Service in 2002"

QoS-driven server migration for Internet data centers

Abstract: Many organizations have chosen to host Internet applications at Internet data centers (IDCs) located near network access points of the Internet to take advantage of their high availability, large network bandwidths and low network latencies. Current IDCs provide for a dedicated and static allocation of resources to each hosted application. Unfortunately, workloads for these sites are highly variable, leading to poor resource utilization, poor application performance, or both. In this paper, we develop a framework for QoS-driven dynamic resource allocation in IDCs. Termed QuID (quality of service infrastructure on demand), the framework's contributions are threefold. First, we develop a simple adaptive algorithm to reduce the average number of servers used by an application while satisfying its QoS objectives. Second, we develop an optimal off-line algorithm that bounds the advantage of any dynamic policy and provides a benchmark for performance evaluation. Finally, we perform an extensive simulation study using traces from large-scale E-commerce and search-engine sites. We explore the gains of the QuID algorithms as a function of the system parameters (such as server migration time), algorithm parameters (such as control time scale), and workload characteristics (such as peak-to-mean ratio and autocorrelation function of the request rate).

Achieving efficient channel utilization and weighted fairness for data communications in IEEE 802.11 WLAN under the DCF

Abstract: Fair allocation of bandwidth and maximization of channel utilization are two important issues when designing a contention-based wireless medium access control (MAC) protocol. However, achieving both design goals at the same time is very difficult, and has not yet been addressed elsewhere. We study this challenging problem, particularly for data communications in IEEE 802.11 wireless local area networks (WLANs). We propose a priority-based fair medium access control (P-MAC) protocol by modifying the distributed coordination function (DCF) of the IEEE 802.11 MAC. The key idea is that the contention window size for each wireless station is properly selected to reflect: (1) the relative weights among data traffic flows, so as to achieve the weighted fairness; (2) the number of stations contending for the wireless medium, so as to maximize the aggregate throughput. In P-MAC, our approximations to the optimal contention window sizes, which are based on a theoretical analysis, are evaluated numerically and shown to work well under different network configurations and traffic scenarios. Moreover, simulation results show that, with few changes to the original DCF, P-MAC performs significantly better in terms of both fairness and throughput.

Quality of service scheduling in cable and broadband wireless access systems

Abstract: Several last mile high-speed technologies have been explored to provide Internet access and multimedia services to end users. Most notable of those technologies are hybrid fiber coax (HFC) cable networks, digital subscriber line (DSL), satellite access, and fixed broadband wireless access (BWA) systems. The de facto standard for delivering broadband services over HFC networks is the Data Over Cable Service Interface Specifications (DOCSIS) protocol. For BWA systems, on the other hand, a new protocol, called IEEE 802.16, was developed for the same purpose. This paper presents a new and efficient scheduling architecture to support bandwidth and delay quality of service (QoS) guarantees for both DOCSIS and IEEE 802.16. Our design goals are simplicity and optimum network performance. The architecture we develop supports various types of traffic including constant bit rate, variable bit rate (real-time and non-realtime) and best effort.

An adaptive control framework for QoS guarantees and its application to differentiated caching

Abstract: Software mechanisms that enforce QoS guarantees often require knowledge of platform capacity and resource demand. This requirement calls for performance measurements and profiling upon platform upgrades, failures, or new installations. The cost of performing such measurements is a significant hurdle to the wide-spread deployment of open QoS-aware software components. In this paper, we introduce a new QoS-control paradigm based on adaptive control theory. The hallmark of this paradigm is that it eliminates profiling and configuration costs of QoS-aware software, by completely automating the process in a way that does not require user intervention. As a case study, we describe, implement and evaluate the control architecture in a proxy cache to provide proportional differentiation on content hit rate. Adaptive control theory is leveraged to manage cache resources in a way that adjusts the quality spacing between classes, independently of the class loads, which cannot be achieved by other cache resource management schemes, such as biased replacement policies, LRV or greedy-dual-size.

Defending against distributed denial-of-service attacks with max-min fair server-centric router throttles

Abstract: We present a network architecture and accompanying algorithms for countering distributed denial-of-service (DDoS) attacks directed at an Internet server. The basic mechanism is for a server under stress to install a router throttle at selected upstream routers. The throttle can be the leaky-bucket rate at which a router can forward packets destined for the server. Hence, before aggressive packets can converge to overwhelm the server, participating routers proactively regulate the contributing packet rates to more moderate levels, thus forestalling an impending attack. In allocating the server capacity among the routers, we propose a notion of level-k max-min fairness. We present a control-theoretic model to evaluate algorithm convergence under a variety of system parameters. In addition, we present packet network simulation results using a realistic global network topology, and various models of good user and attacker distributions and behavior. Using a generator model of Web requests parameterized by empirical data, we also evaluate the impact of throttling in protecting user access to a Web server. First, for aggressive attackers, the throttle mechanism is highly effective in preferentially dropping attacker traffic over good user traffic. In particular, level-k max-min fairness gives better good-user protection than recursive pushback of max-min fair rate limits proposed in the literature. Second, throttling can regulate the experienced server load to below its design limit - in the presence of user dynamics - so that the server can remain operational during a DDoS attack.

An observation-based approach towards self-managing Web servers

Abstract: The Web server architectures that provide performance isolation, service differentiation, and QoS guarantees rely on external administrators to set the right parameter values for the desired performance Due to the complexity of handling varying workloads and bottleneck resources, configuring such parameters optimally becomes a challenge In this paper we describe an observation-based approach for self-managing Web servers that can adapt to changing workloads while maintaining the QoS requirements of different classes In this approach, the system state is monitored continuously and parameter values of various system resources-primarily the accept queue and the CPU-are adjusted to maintain the system-wide QoS goals We implement our techniques using the Apache Web server and the Linux operating system We first demonstrate the need to manage different resources in the system depending on the workload characteristics We then experimentally demonstrate that our observation-based system can adapt to workload changes by dynamically adjusting the resource shares in order to maintain the QoS goals

Mitigation of DoS attacks through QoS regulation

Abstract: As more and more critical services are provided over the Internet, the risk to these services from malicious users is also increasing. Several networks have witnessed denial of service attacks in the past. This paper reports on our experience in building a Linux-based prototype to mitigate the effect of such attacks. Our prototype provides an efficient way to keep track of server and network resources at the network layer and allows aggregate resource regulation. Our scheme provides a general, and not attack specific, mechanism to provide graceful server degradation in the face of such an attack. We report on the rationale of our approach, the experience in building the prototype, and the results from real experiments. We show that traditional rate-based regulation combined with proposed window-based regulation of resources at the aggregate level at the network layer is a feasible vehicle for mitigating the impact of DOS attacks on end servers.

The packet size dependence of packet pair like methods

Abstract: Packet-pair based link estimation methods allow the estimation of bottleneck bandwidth in Internet routes. In practice, several complicating effects combine which can seriously distort such estimates. We provide a new delay variation based route model which allows the principles of packet-pair to be formalised and extended. This enables the effect of probe size to be evaluated, downstream noise understood, peak detection recognised as superior to mode or minimum based filtering, and new estimation methods to be proposed and evaluated. Using insight from the governing equations and simulation, it is shown how real measurements made over a 12 hop route can be interpreted. Unexpected additional probe size dependencies were found, inspiring an extension of the route model to include lower layer headers. It is shown how the enhanced model accounts very well for the observed dependencies, allowing more accurate estimates and a greater understanding of the role of cross traffic.

Supporting low latency TCP-based media streams

Abstract: The dominance of the TCP protocol on the Internet and its success in maintaining Internet stability has led to several TCP-based stored media-streaming approaches. The success of these approaches raises the question whether TCP can be used for low-latency streaming. Low latency streaming allows responsive control operations for media streaming and can make interactive applications feasible. We examined adapting the TCP send buffer size based on TCP's congestion window to reduce application perceived network latency. Our results show that this simple idea significantly improves the number of packets that can be delivered within 200 ms and 500 ms thresholds.

QoS-aware component frameworks

Abstract: Developers of QoS (quality of service) applications need component frameworks that support QoS to improve their development process. We propose to use the component infrastructure level for the integration of QoS facilities to avoid the problems of QoS infrastructures dependencies, and to simplify the development process of applications. This paper introduces our model of QoS-aware component, and describes its relevant entities. In our support of the model, the component descriptor provides facilities for the configuration of QoS attributes, and the component container isolates the QoS business component from the basic QoS infrastructures, which are supported in QoS-aware networks, OS and middleware. In this paper we include some practical results based on QoS-aware components that include support to guarantee the response times, jitter and resource consumption based on resource reservation services.

Aliquem: a novel DRR implementation to achieve better latency and fairness at O(1) complexity

Abstract: Deficit round-robin is a packet scheduling algorithm devised to provide fair queueing in the presence of variable length packets (see Shreedhar, M. and Varghese, G., IEEE Trans. on Networking, vol.4, p.375-85, 1996). DRR can be implemented at O(1) complexity provided that each flow is allowed to transmit at least one maximum size packet on each round; however, under this constraint, DRR may exhibit high latency and poor fairness. We first generalize previous results known for DRR, related to its latency and fairness. We then introduce a novel DRR implementation technique, called active lists queue method (Aliquem), which allows the above constraint to be relaxed while preserving O(1) complexity, thus achieving better latency and fairness that are comparable to those of more complex algorithms, such as self-clocked fair queueing.

A SLA framework for QoS provisioning and dynamic capacity allocation

Abstract: Any QoS scheme must be designed from the perspective of pricing policies and service level agreements (SLAs). Although there has been enormous amount of research in designing mechanisms for delivering QoS, its applications has been limited due to the missing link between QoS, SLA and pricing. Therefore the pricing policies in practice are very simplistic (fixed price per unit capacity with fixed capacity allocation or pricing based on peak or 95-percentile load etc.). The corresponding SLAs also provide very limited QoS options. This leads to provisioning based on peak load, under-utilization of resources and high costs. We present a SLA based framework for QoS provisioning and dynamic capacity allocation. The proposed SLA allows users to buy a long term capacity at a pre-specified price. However, the user may dynamically change the capacity allocation based on the instantaneous demand. We propose a three tier pricing model with penalties (TTPP) SLA that gives incentives to the users to relinquish unused capacities and acquire more capacity as needed. This work may be viewed as a pragmatic first step towards a more dynamic pricing scenario. We solve the admission control problem arising in this scheme using the concept of trunk reservation. We also show how the SLA can be used in virtual leased-line service for VPNs, and Web hosting service by application service providers (ASPs). Using Web traces we demonstrate the proposed SLA can lead to more efficient usage of network capacity by a factor of 1.5 to 2. We show how this translates to payoffs to the user and the service provider.

A graph theoretic approach to bounding delay in proxy-assisted, end-system multicast

Abstract: End-system multicast provides a low-cost solution to scalably broadcast information to groups of users. However, last-mile bandwidth limitations constrain tree fanouts; leading to high end-to-end delivery delays. These delays can be reduced if the network provides forwarding proxies with high fanout capabilities at an additional cost. We use simple graph theoretic network models to explore the problem of building hybrid proxy/end-system application layer multicast trees that meet fixed end-to-end delay bounds. Our goal is to meet a fixed delay bound while minimizing costs associated with the utilization of proxies. We provide an algorithm and formally prove its optimality in a fully-connected overlay network with uniform-length edges. We then adapt this algorithm into a heuristic and evaluate the heuristic for simulated transit-stub networks with variable-delay edges. We compare our heuristic in a proxy-free environment to previously developed heuristics and show that our heuristic typically yields further reductions in the maximum session end-to-end delay.

Assured and expedited forwarding extensions for IEEE 802.11 wireless LAN

Abstract: We propose DIME (DiffServ MAC extension), an extension of the IEEE 802.11 MAC protocol to support differentiated services. The proposed extension consists of two optional modules: expedited forwarding (EF) and assured forwarding (AF). The expedited forwarding extension (DIME-EF) reuses the interframe space of the point coordination function (PCF) of the IEEE 802.11 standard in a distributed manner, while the assured forwarding extension (DIME-AF) relies on the distributed coordination function (DCF) with a modified algorithm for the computation of the contention window (CW). Best effort is supported by the functionality of the current 802.11 standard in such a way that legacy IEEE 802.11 terminals behave as best effort terminals in the DIME architecture. While the performance of the assured forwarding extension has been thoroughly evaluated by the authors elsewhere (see Banchs, A. and Perez, X., Proc. IEEE Wireless Commun. and Networking Conf., 2002), this paper concentrates on the overall architecture and the performance of the expedited forwarding extension.

Speech recognition performance as an effective perceived quality predictor

Abstract: Determining the perceived quality of packet audio under packet loss usually requires human-based mean opinion score (MOS) listening tests. We propose a new MOS estimation method based on machine speech recognition. Its automated, machine-based nature facilitates real-time monitoring of transmission quality without the need to conduct time-consuming listening tests. Our evaluation of this new method shows that it can use the word recognition ratio metric to reliably predict perceived quality. In particular, we find that although the absolute word recognition ratio of a speech recognizer may vary depending on the speaker, the relative word recognition ratio, obtained by dividing the absolute word recognition ratio with its own value at 0% loss, is speaker-independent. Therefore the relative word recognition ratio is well suited as a universal, speaker-independent MOS predictor. Finally we have also conducted human-based word recognition tests and examined its relationship with machine-based recognition results. Our analysis shows that they are correlated although not very linearly. Also we find that human-based word recognition ratio does not degrade significantly once packet loss is large (/spl ges/10%).

A scalable and robust solution for bandwidth allocation

Abstract: We propose a novel architecture for providing bandwidth allocation and reservation that is both scalable and robust. Scalability is achieved by not requiring routers to maintain per-flow state on either the data or control planes. To achieve robustness, we develop two key techniques. First, we use an admission control mechanism based on lightweight certificates and random sampling to prevent malicious users from claiming reservations that were never allocated to them. Second, we use a recursive monitoring algorithm to detect misbehaving flows that exceed their reservations. We randomly divide the traffic into large aggregates, and then compare the data arrival rate of each aggregate to its reservation. If an aggregate misbehaves, i.e., its arrival rate is greater than its reservation, we split and monitor that aggregate recursively until we detect the misbehaving flow(s). These misbehaving flows are then policed separately. We conduct extensive simulations to evaluate our solutions. The results show that the proposed solution is very effective in protecting well-behaved flows when the fraction of misbehaving flows is limited.

Probe-based admission control for multicast

Abstract: End-to-end measurement based admission controls (MBAC) have recently been proposed to support quality of service for real-time transfer of data. All these designs share the idea of decentralizing the admission decision by requiring each end host to probe the network before transmission. These schemes are solely targeted at unicast communications, while multicast data has not yet been addressed. We study a probing procedure to perform admission decisions for multicast senders and receivers. The admission control offers a reliable upper bound on the packet loss for the multicast session even with short probe phase durations (e.g. half a second). Our probing mechanism only requires the routers to differentiate between two classes of packets: high priority data and low priority probes. Simulation results of the performance of the procedure are presented and evaluated.

Robust bandwidth allocation strategies

Abstract: Allocating bandwidth for a certain period of time is an often encountered problem in networks offering some kind of quality of service (QoS) support. In particular, for aggregate demand the required bandwidth at each point in time may exhibit considerable fluctuations, random fluctuations as well as systematic fluctuations due to different activity at different times of day. In any case, there is a considerable amount of uncertainty to be dealt with by strategies for effectively allocating bandwidth. In this paper, we try to devise so-called robust strategies for bandwidth allocation under uncertainty. The notion of robustness here means that we look for strategies which perform well under most circumstances, but not necessarily best for a given situation. By simulations, we compare the different strategies we propose with respect to the robustness and performance they achieve in terms of (virtual) cost savings. We show that robustness and good performance need not be contradictory goals and furthermore that very good strategies need not be complex, either.

A case for tree evolution in QoS multicasting

Abstract: The phenomenal growth of group communications and QoS-aware applications over the Internet have accelerated the development of multicasting technologies. The core-based tree (CBT) multicasting approach provides a scalable solution for large groups for large networks such as the Internet. However, unlike in shortest-path trees, the quality (tree cost) of the CBT may eventually degrade over time due to group dynamics (join/leave). In order to counteract this degradation, the core may be migrated and a new tree constructed. The method of migrating group members from the old core to the new core has a profound impact on the quality of the tree and also on the service disruption experienced by group members. Thus, there exists a trade-off between tree cost and service disruption as higher rate of migration decreases the overall tree cost but results in more service disruption. Chakrabarti and Manimaran (see Proc. IEEE Globecom, 2001) developed the concept of tree migration. In this paper, we develop a new paradigm for tree migration, namely tree evolution. The proposed tree evolution model (Split-based Tree Evolution Protocol (STEP)) provides an elegant solution that strikes a balance between service disruption and tree cost for highly dynamic groups. We compare and contrast the merits of tree evolution versus tree migration. Our simulation studies show that the proposed evolution model demonstrates excellent tree cost and service disruption for highly dynamic groups.

Simplification of network dynamics in large systems

Abstract: We show that significant simplicity can be exploited for pricing-based control of large networks. We first consider a general loss network with Poisson arrivals and arbitrary holding time distributions. In dynamic pricing schemes, the network provider can charge different prices to the user according to the current utilization level of the network and other factors. We show that, when the system becomes large, the performance (in terms of expected revenue) of an appropriately chosen static pricing scheme, whose price is independent of the current network utilization, approaches that of the optimal dynamic pricing scheme. Further, we show that, under certain conditions, this static price is independent of the route the flows take. This indicates that we can use the static scheme, with its much simpler structure, to control large communication networks. We then extend the result to the case of dynamic routing, and show that the performance of an appropriately chosen static pricing scheme, with bifurcation probability determined by average parameters, can also approach that of the optimal dynamic routing scheme when the system is large. Finally, we study the control of elastic flows and show that there exist schemes with static parameters whose performance can approach that of the optimal dynamic resource allocation scheme (in the large system limit). We also identify the applications of our results to QoS routing and rate control for real-time streaming.

Experiences with evaluating network QoS for IP telephony

Abstract: Successful deployment of networked multimedia applications such as IP telephony depends on the performance of the underlying data network. QoS requirements of these applications are different from those of traditional data applications. For example, while IP telephony is very sensitive to delay and jitter, traditional data applications are more tolerant of these performance metrics. Consequently, assessing a network to determine whether it can accommodate the stringent QoS requirements of IP telephony becomes critical. We describe a technique for evaluating a network for IP telephony readiness. Our technique relies on the data collection and analysis support of our prototype tool, ExamiNet/spl trade/. It automatically discovers the topology of a given network and collects and integrates network device performance and voice quality metrics. We report the results of assessing the IP telephony readiness of a real network of 31 network devices (routers/switches) and 23 hosts via ExamiNet/spl trade/. Our evaluation identified links in the network that were over utilized to the point at which they could not handle IP telephony.

Performance evaluation of a QoS-aware framework for providing tunable consistency and timeliness

Abstract: Strong replica consistency models ensure that the data delivered by a replica always includes the latest updates, although this may result in poor response times. On the other hand, weak replica consistency models provide quicker access to information, but do not usually provide guarantees about the degree of staleness in the data they deliver. In order to support emerging distributed applications that are characterized by high concurrency demands, an increasing shift towards dynamic content, and timely delivery, we need quality of service models that allow us to explore the intermediate space between these two extreme approaches to replica consistency. Further, for better support of time-sensitive applications that can tolerate relaxed consistency in exchange for better responsiveness, we need to understand how the desired level of consistency affects the timeliness of a response. The QoS model we have developed to realize these objectives considers both timeliness and consistency, and treats consistency along two dimensions: order and staleness. We evaluate experimentally the framework we have developed to study the timeliness/consistency tradeoffs for replicated services and present experimental results that compare these tradeoffs in the context of sequential and FIFO ordering.

End-to-end delay guarantees for multiple-channel schedulers

Abstract: Consider a network in which adjacent nodes exchange messages via multiple communication channels. Multiple channels between adjacent nodes are desirable due to their cost effectiveness and improved fault-tolerance. We consider the problem of providing deterministic quality of service guarantees in this network. We show that any scheduling protocol designed for a single channel can be converted into a multiple-channel scheduling protocol without significantly increasing the delay at the scheduling node. However, because there are multiple channels between adjacent nodes, the packets of a flow may be reordered. This in turn significantly increases the upper bound on the end-to-end delay of the flow. We show how this increase in delay can be avoided through the use of efficient sorting techniques.

Resource management for IP telephony networks

Abstract: The two key resources in an IP telephony network are the Internet telephony gateways (ITGs) and the IP network. These resources must be effectively managed to simultaneously provide good QoS to calls and maximize network resource utilization. This paper presents two main contributions. First, we design a call admission policy based on congestion sensitive pricing. As the load increases, this policy preferentially admits users who place a higher value on making a call while simultaneously maintaining a high utilization of network resources. We derive the function mapping congestion to price for the admission policy that maximizes revenue. Second, we design a call redirection policy to select the best ITG to serve the call. The policy balances load to improve network efficiency and incorporates QoS sensitivity to improve call quality. Simulation results show the following: (i) congestion pricing based admission control lowers call blocking probability, increases provider revenue, and improves economic efficiency over a static flat-rate admission control scheme; (ii) congestion sensitivity in the redirection policy balances load across all the ITGs while QoS sensitivity improves call audio quality; and (iii) incorporating price sensitivity in the redirection policy improves the economic efficiency, i.e., ensures that users who pay more get higher QoS. The techniques studied in this paper can be combined into a single resource management solution that can improve network resource utilization, provide differentiated service, and maximize provider revenue.

Operating system support for network control: a virtual network interface approach for end-host OSs

Abstract: Because of user demands for better quality of service, network-aware applications have been of increasing necessity. To enable more control, the end-host operating system (OS) is the entity responsible for providing appropriate service level and API to user applications. However, most of the work in this area remains domain-specific and without a generalizable scheme for providing network control as an OS service. We propose an OS service, namely the virtualization of network interface, that lies between network interface and userland. The virtual network interface is hierarchically attachable to various OS-supported entity, such as threads, processes, and sockets. We argue that the mechanism provides flexible control, as well as the system protection that is required for operating system services. For a proof of the concept, we show an implementation on a PC-Unix, using the procfs file system abstraction. We also carried out a systematic evaluation. The system exhibited the expected control behavior, while keeping the performance small.

A QoS framework for clusters to support applications with resource adaptivity and predictable performance

Abstract: Simultaneous advances in processor and network technologies have made clusters of workstations attractive vehicles for high-performance computing. Emerging applications targeted for clusters are inherently interactive and collaborative in nature. These applications demand end-to-end quality of service (QoS) in addition to performance. Achieving predictable performance and ability to exploit resource adaptivity are also common requirements of these next generation applications. Providing QoS mechanisms for clusters to satisfy the demands of next generation applications is a challenging task. We propose a QoS framework that provides bandwidth guarantees for communication within a cluster. The framework consists of a novel network interface card (NIC)-based rate control mechanism and a coordinated admission control/scheduling mechanism. An interface is developed so that applications using the common message passing interface (MPI) standard can specify bandwidth requirements of their flows to the underlying network. The framework is developed and evaluated on a Myrinet cluster testbed for a range of scientific and visualization applications. The experimental evaluations demonstrate the various advantages such as predictability and resource adaptability associated with the framework. The proposed framework is quite unique and is the first of its kind in the literature to support next generation interactive and collaborative applications on clusters.

The performance of GPS and EDF with temporary sessions

Abstract: We study the behavior of packet scheduling protocols in a temporary sessions model in which sessions come and go over time. We first show that, in this setting, the standard delay bound for the well-studied generalized processor sharing (GPS) protocol can be exceeded by an arbitrary amount. In extreme cases, GPS can actually be unstable. In contrast, we then show that it is possible to choose deadlines so that the earliest-deadline-first (EDF) protocol is always stable. We illustrate our results by analysis and simulations.