Mobile Edge Computing (MEC) enables both to prolong the battery life of mobile devices and support the execution of computationally intensive applications at the edge. This can be achieved by offloading these tasks to a server deployed near the base station and/or by directly caching them. Previous works focus on only one of these two strategies or formulate optimization problems that are hard to solve and propose a suboptimal solution. In this paper, we propose a linear model for the joint task caching and offloading optimization problem. Moreover, we present two efficient heuristics which provide close-to-optimal results in terms of energy efficiency with a low execution time. We further prove that the offloading subproblem can be solved with an optimal algorithm. Finally, we demonstrate the performance and scalability of our propositions by extensive simulations on a large number of $10^{5}$ mobile devices

Video streaming dominates the Internet traffic. Assessing the carbon footprint of video streaming has received recently a significant attention with a number of models proposed to associate a CO2cost to one hour of streaming. In this work, we compare the modeling assumptions and computation methods used by five recent works to inform the debate. Indeed, initial results can be at odds, with up to one order of magnitude difference in the estimates. Our contributions are: (i) we relate the difference in the results primarily to the perimeter of the study, e.g. including production cost or not, (ii) we question some of the modeling assumptions made using a real deployment of a streaming server in a controlled environment with up to 2000 clients and (iii) we propose a technique to reconcile the models and obtain a CO2estimate in between 60 and $\mathbf{1 4 0}$ grams when considering the average worldwide carbon intensity of electricity.

In the last years, real-time media transport using QUIC has aroused general interest. QUIC now features a datagram mode (congestion control but no loss recovery) along the legacy stream mode. Ongoing research is studying protocol mechanisms to transport media with QUIC to build a new streaming protocol or to map existing ones like RTP onto QUIC. Our work focuses on the transport of RTP packets generated by WebRTC over QUIC and investigates the resulting in-teraction of the QUIC and WebRTC congestion control algorithms. We performed extensive experiments to study various combinations of congestion control algorithms at the QUIC and WebRTC levels. We observed that in a low latency setup, e.g. FTTH, using the stream mode of QUIC pays off as it hides losses to WebRTC, and hence maintaining the maximum video quality otherwise lost. For higher latency scenarios, the datagram mode with an appropriate QUIC level transport pro-tocol offers good performance, even though in case of severe changes of the channel conditions, the WebRTC mechanism takes control of the situation by tuning the encoder rate.

Despite the rapid advancements in 5G technology, accurately assessing the energy consumption of its Radio Ac-cess Networks (RANs) remains a challenge due to the diverse range of applicable technologies and implementation solutions. Designing a versatile power model for estimating the 5G RAN-specific power consumption requires extensive data collection and experimental studies to capture the diverse range of technolo-gies and implementation solutions. The objective is to outline a versatile energy model capable of estimating RAN-specific energy consumption, encompassing both mobile terminals and the physical layer (PHY) of base stations. In this paper, we focus on the computational complexity of the baseband part of the model. The developed (part of the) model is compared with the estimation of the number of cycles (and energy per cycle) used by a specific implementation (here a Matlab code ported on an Intel target), enabling the assessment of the model with the estimation of energy consumed on a real target. The study's results show a good agreement between the model and the implementation, even if some parts need to be refined to take specific algorithms into account. The key contribution is the development of an initial flexible energy model with finer granularity, enabling comparisons of energy use across various applications and contexts, and offering a comprehensive tool for optimizing 5G network energy consumption.

In the last years, Real-Time media transport using QUIC has aroused general interest. Ongoing research is studying protocol mechanisms to transport media with QUIC to build a new streaming protocol or to map existing ones like RTP onto QUIC. Our work focuses on the transport of RTP packets generated by WebRTC over QUIC and investigates the combination of the QUIC and WebRTC congestion control algorithms. To do so, we devised a testbed to study various combinations of congestion control algorithms, for different QUIC implementations, when sending RTP packets from an un-modified WebRTC client (Chrome).

A hybrid CDN/Viewer-to-Viewer (V2V) architecture is an attractive solution for HTTP (HLS) and MPEG-DASH-based live streaming providers. It combines a traditional CDN with a V2V overlay for exchanging video fragments, reducing the cost of the CDN while maintaining the quality of experience. This work explores machine learning models to address the key challenge of neighbor selection. Our goal is to predict the connection quality between two arbitrary viewers using features such as locality, access providers, operating systems, past CDN, and V2V throughput. The proposed solutions are validated using an A/B testing approach on our production system, demonstrating a significant improvement in key system metrics compared to the traditional locality-based methods. We observe 17% higher V2V throughput, 26% lower delay, 37% fewer lost chunks, 39% fewer re-buffering, and 20% fewer quality switches.

Maintaining up-to-date KVM hypervisors requires regular upgrades to the host kernel, hence rebooting the physical host with the consequent termination of running Virtual Machines (VMs). Cloud platforms capable of massive large-scale live migrations evacuate VMs from the hosts before rebooting, minimizing the impact over VM up-time. However, scenarios exist where resource constraints make live migration undesirable, or the presence of fault-tolerant instances (e.g., replicated services) favors the adoption of VM termination, a simpler but more disruptive strategy. In this article, we present Hy-FiX, a fast in-place upgrade solution for KVM hypervisors. Hy-FiX preserves VM memory across host reboots, protecting the execution state of running guests while hypervisor upgrades are applied. Hy-FiX memory preservation across reboot, combined with a mixed suspend-to-disk/suspend-to-RAM technique, achieves a 2.31-second checkpoint/restore time for a 256 GB VM, and Hy-FiX lazy memory initialization reboots an enterprise-class host in constant time (7.6 seconds) regardless of its equipped memory. Hy-FiX is, therefore, a better alternative to classical VM termination and restart.

The importance of Big Data is nowadays established, both in industry and research fields, especially stream processing for its capability to analyze continuous data streams and provide statistics in real-time. Several data stream processing (DSP) platforms exist like the Storm, Flink, Spark Streaming and Heron Apache projects, or industrial products such as Google MillWheel. Usually, each platform is tested and analyzed using either specifically crafted benchmarks or realistic applications. Unfortunately, these applications are only briefly described and their source code is generally not available. Hence, making quick evaluations often involves rewriting complete applications on different platforms. The lack of a generic prototype application also makes it difficult for a developer to quickly evaluate the impact of some design choices. To address these issues, we present NAMB (Not only A Micro-Benchmark), a generic application prototype generator for DSP platforms. Given a high-level description of a stream processing application and its workload, NAMB automatically generates the code for different platforms. It features a flexible architecture which makes it easy to support new platforms. We demonstrate the benefits of our proposal to quickly generate application prototypes as well as benchmarks used in published papers. Overall, our approach provides easily replicable, comparable and customizable prototypes for data stream platforms. Moreover, NAMB provides similar performance in terms of latency and throughput to existing benchmarks, while only requiring a simple high-level description.

Live video distribution represents a significant fraction of the video traffic in the Internet. Its distribution induces heavy costs to serve clients during peak hours. To tackle this issue, the hybrid Viewer-to-Viewer (V2V) mode - where peers cooperate with each other in addition to receiving data from a Content Distribution Network (CDN) - has been proposed to both improve the Quality of Experience (QoE) and reduce the distribution cost. In this paper, we present the improvements we have introduced in a commercial V2V Web live streaming system based on the Web Real-Time Communication (WebRTC) standard. A key challenge faced when designing such a system is the algorithm that governs the client's behavior and should maximize the amount of a data exchanged in V2V mode. A crucial component in V2V Web live streaming is the fast distribution of the video chunks availability in the V2V network. We present in this work our initial reactive approach to distribute the information concerning chunks availability and contrast it with our new proactive approach. We demonstrate the efficiency of the proactive approach on a testbed featuring light clients (browsers) separated using Linux network namespaces, as well as with live experiments. The fraction of V2V traffic for the production environment also improved from 19% in the reactive approach to 37% in the proactive one. Furthermore, we demonstrate that the buffer length of the clients both in the controlled environment and in production is improved, implying a better QoE.

The relevance of Data Stream Processing (DSP) is nowadays established, thanks to its capability to analyze continuous streams and provide statistics in real-time. A considerable amount of work has been dedicated to improve performance and features of DSP platforms. Thus, benchmark application are necessary for comparison and evaluation. Unfortunately, in literature, these applications are often briefly described, the source is not available, they are too context-specific or don't provide enough flexibility. That makes it difficult for a developer to quickly evaluate the impact of some design choices. To address these issues, we introduce a high-level description model of stream applications. Based on fundamental DSP characteristics, this description allow an easy and flexible definition of benchmark topologies. With this model we aim to provide easily replicable, comparable and customizable benchmarks for DSP. We then use a framework prototype that translates the high-level description into platform-specific code simulating the application workload.

The widespread use of social networks and applications such as IoT networks generates a continuous stream of data that companies and researchers want to process, ideally in real-time. Data stream processing systems (DSP) enable such continuous data analysis by implementing the set of operations to be performed on the stream as directed acyclic graph (DAG) of tasks. While these DSP systems embed mechanisms to ensure fault tolerance and message reliability, only few studies focus on the impact of these mechanisms on the performance of applications at runtime. In this paper, we demonstrate the impact of the message reliability mechanism on the performance of the application. We use an experimental approach, using the Storm middleware, to study an acknowledgment-based framework. We compare the two standard schedulers available in Storm with applications of various degrees of parallelism, over single and multi cluster scenarios. We show that the acking layer may create an unforeseen bottleneck due to the acking tasks placement; a problem which, to the best of our knowledge, has been overlooked in the scientific and technical literature. We propose two strategies for improving the acking tasks placement and demonstrate their benefit in terms of throughput and latency.

Hypervisors are critical components in cloud data centers. Maintaining up-to-date KVM hypervisors requires replacing the host kernel, hence rebooting the physical host with consequent termination of running virtual machines (VMs). We introduce Hy-FiX, an in-place upgrade tool for KVM hypervisors that applies host kernel and/or user-space components upgrades while transparently preserving running VMs, constituting a better alternative for platforms where VMs are inevitably terminated and restarted.

Idle virtual machines (VMs) are a waste of resources in data centers. We introduce SEaMLESS, which transforms a fully-Hedged idle VM into a lightweight and resourceless Virtual Network Function (VNF). Idle VMs can then be saved to disk and release their memory. Simultaneously, the VNF provides service availability. Upon user activity, the appropriate VM is restored, without introducing any interruption for service users. Tens of VNFs can be contained within the same memory space required for one single VM, thereby facilitating ample resources savings when scaled up to a data center.

Energy aware routing aims at reducing the energy consumption of ISP networks. The idea is to adapt routing to the traffic load in order to turn off some hardware. However, it implies to make dynamic changes to routing configurations which is almost impossible with legacy protocols. The Software Defined Network (SDN) paradigm bears the promise of allowing a dynamic optimization with its centralized controller. In this work, we propose SENAtoR, an algorithm to enable energy aware routing in a scenario of progressive migration from legacy to SDN hardware. Since in real life, turning off network equipments is a delicate task as it can lead to packet losses, SENAtoR provides also several features to safely enable energy saving services: tunneling for fast rerouting, smooth node disabling and detection of both traffic spikes and link failures. We validate our solution by extensive simulations and by experimentation. We show that SENAtoR can be progressively deployed in a network using the SDN paradigm. It allows to reduce the energy consumption of ISP networks by 5 to 35% depending on the penetration of SDN hardware, while diminishing the packet loss rate compared to legacy protocols.

Video traffic constitutes the majority of traffic in bytes that mobile and fixed line operators deliver to their customer. This type of traffic is both resource consuming and QoE sensitive. Either because of content quality or QoE, a large fraction of users often abandon viewing prematurely. These abandonment phenomena lead to a huge waste of network resources and device batteries. Several strategies have been devised to account for all those dimensions. Dominant approaches are fast-caching where the server pushes traffic as fast as possible to the client in order to limit starvation, and ON-OFF strategy where the client forces the server to pause the transfer regularly in order to mitigate the wasted bytes and energy due to users abandonment. In this work, we focus on fast-caching and on ON-OFF strategies. We develop an analytical model which takes into account users dynamics and allows us to quantify the loss in bytes due to users' viewing abandonment. Furthermore, we formulate a multi-objective optimization problem to find ON and OFF period durations that feature a good trade-off between loss due to abandonment and starvation probability.

Software Defined Networking (SDN) is gaining momentum with the support of major manufacturers. While it brings flexibility in the management of flows within the data center fabric, this flexibility comes at the cost of smaller routing table capacities. In this paper, we investigate compression techniques to reduce the forwarding information base (FIB) of SDN switches. We validate our algorithm, called MINNIE, on a real testbed able to emulate a 20 switches fat tree architecture. We demonstrate that even with a small number of clients, the limit in terms of number of rules is reached if no compression is performed, increasing the delay of all new incoming flows. MINNIE, on the other hand, reduces drastically the number of rules that need to be stored with a limited impact on the packet loss rate. We also evaluate the actual switching and reconfiguration times and the delay introduced by the communications with the controller.

In the last decade, path discovery has been extensively covered in the literature. In its simplest form, it generally works by sending probes that expire along the path from a host to a destination. It is also known that network administrators often configure their routers to limit the amount of ICMP replies sent, a common practice typically referred to as ICMP rate limitation. In this paper we attempt to characterize the responsiveness of routers to expiring ICMP echo-request packets. Our contribution is twofold: first, we provide a detailed analysis of how routers are most commonly configured to respond to expiring packets; next, we show that for the vast majority of routers, the measured round-trip time is not affected by the probing rate.

Software-Defined Networking (SDN) enables consolidation of the control plane of a set of network equipments with a fine-grained control of traffic flows inside the network. In this work, we demonstrate that some coarse-grained scheduling mechanisms can be easily offered by SDN switches without requiring any unsupported operation in OpenFlow. We leverage the feedback loop - flow statistics - exposed by SDN switches to the controller, combined with priority queuing mechanisms, usually available in typical switches on their output ports. We illustrate our approach through experimentations with an OpenvSwitch SDN switch controlled by a Beacon controller.

BGP hijacking is a well known threat to the Internet routing infrastructure. There has been considerable interest in developing tools that detect prefix hijacking but such systems usually identify a large number of events, many of them being due to some benign BGP engineering practice or misconfiguration. Ramachandran et al. [1] and later Hu et al. [2] also correlated suspicious routing events with spam and claimed to have found evidence of spammers temporarily stealing prefixes to send spam. In an effort to study at large scale the existence and the prevalence of malicious BGP hijacks in the Internet we developed a system which (i) identifies hijacks using BGP, traceroute and IRR data and (ii) investigates traffic originating from the reported networks with spam and netflow data. In this paper we present a real case where suspicious BGP announcements coincided with spam and web scam traffic from corresponding networks. Through this case study we show that a correlation of suspicious routing events with malicious activities is insufficient to evidence harmful BGP hijacks. We thus question previously reported cases and conclude that identifying malicious BGP hijacks requires additional data sources as well as feedback from network owners in order to reach decisive conclusions.

While Internet traffic has received a lot of attention, little is known about enterprise traffic. In this paper, we shed light on the basic characteristics of modern enterprise networks using a 24-hour long packet trace captured in a mid-size academic network. Our approach is to contrast the external and internal activities of the enterprise network under study. We demonstrate that these two types of traffic differ along several dimensions of our analysis, especially traffic composition, symmetry level or RTT. Overall, such a study provides valuable insights for traffic engineering in enterprise networks and could pave the way towards the design of specific workload models for enterprise networks.

Data traffic in cellular networks increased tremendously over the past few years and this growth is predicted to continue over the next few years. Due to differences in access technology and user behavior, the characteristics of cellular traffic can differ from existing results for wireline traffic. In this study we focus on understanding the flow rates and on the relationship between the rates and other flow properties by analyzing packet level traces collected in a large cellular network. To understand the limiting factors of the flow rates, we further analyze the underlying causes behind the observed rates, e.g., network congestion, access link or end host configuration. Our study extends other related work by conducting the analysis from a unique dimension, the comparison with traffic in wired networks, to reveal the unique properties of cellular traffic. We find that they differ in variability and in the dominant rate limiting factors.

Size-based scheduling improves data transfer response times by favoring flows at an early stage. Although appealing, these techniques raise concerns as they require to keep track of the volume of data sent by each and every ongoing connections and they may starve long-lived flows even if they use up limited bandwidth. Early Flow Discard (EFD) scheduling addresses these issues and we present its adaptation to infrastructure 802.11 networks where the access point downlink queue naturally builds up. To deal with this problem, EFD needs to take into account bi-directional traffic, so that it effectively controls uploads and downloads even though EFD applies to the downlink buffer only. It appears that even with limited buffers, which translates into limited memory of flows for EFD, the most simple flavor of bidirectional EFD -a simple pair of FIFO queues and tracking flow transferred volumes with a packet granularity-enables to rip the full benefit of size-based scheduling, without any of the aforementioned drawbacks.

A host of methods and algorithms have been proposed to solve the issue of traffic classification in recent years. However, a comparison of results between different studies is very difficult due to the lack of structure and common understanding of notions in the domain, especially a precise definition of application classes. This paper aims to fill this gap and propose a first attempt to systematically classify traffic definitions. To attain this goal, we take advantage of the ontology paradigm.

Enterprise networks have a complexity that sometimes rival the one of the larger Internet. Still, enterprise traffic has received little attention so far from the research community. Most studies rely on port numbers to identify applications.

Wireless LANs suffer from performance problems caused by insufficient medium access opportunity given to the access point. Consequently, the downlink buffer fills up, which often leads to packet losses. We propose to address this problem by using a size-based scheduling approach, which is known to favor short flows and the start up of new ones-a very appealing property from the user's perspective as interactive applications and new flows are serviced quickly. Still, size-based scheduling policies have a well-known Achilles heel: large flows can block each other for long periods of time and low rate multimedia transfers may end up with a low priority when their accumulated transferred volume becomes large. To solve the above deficiencies, we propose a new packet scheduling scheme called Least Attained Recent Service (LARS) that applies a temporal decay to the volume of data associated with each flow. In this way, its priority depends more on what has happened recently. With this strategy, LARS can bound the impact of a new arriving flow on ongoing flows, thus limiting lock out durations. It can also efficiently protect low rate multimedia transfers irrespectively of the load conditions.

Accurate identification of network traffic according to application type is a key issue for most companies, including ISPs. For example, some companies might want to ban p2p traffic from their network while some ISPs might want to offer additional services based on the application. To classify applications on the fly, most companies rely on deep packet inspection (DPI) solutions. While DPI tools can be accurate, they require constant updates of their signatures database. Recently, several statistical traffic classification methods have been proposed. In this paper, we investigate the use of these methods for an ADSL provider managing many Points of Presence (PoPs). We demonstrate that statistical methods can offer performance similar to the ones of DPI tools when the classifier is trained for a specific site. It can also complement existing DPI techniques to mine traffic that the DPI solution failed to identify. However, we also demonstrate that, even if a statistical classifier is very accurate on one site, the resulting model cannot be applied directly to other locations. We show that this problem stems from the statistical classifier learning site specific information.

Most tools designed to estimate the capacity of an Internet path require access on both end hosts of the path, which makes them difficult to deploy and use. In this paper we present a single-sided technique for measuring the capacity without the active cooperation of the destination host, focusing particularly on ADSL links. Compared to current methods used on broadband hosts, our approach generates two orders of magnitude less traffic and is much less intrusive. Our tool, DSLprobe, exploits the typical characteristics of ADSL, namely its bandwidth asymmetry and the relatively low absolute bandwidth, in order to measure both downlink and uplink capacities and to mitigate the impact of cross-traffic. To further improve the accuracy, we study different ways to detect and filter cross-traffic packets and we show how to recognize and overcome limited uplink capacities. We validate our tool both on controlled hosts and on a wide variety of Internet hosts. Finally, we present a case study of two large ADSL providers.

In this paper, we investigate the use of a size-based scheduling policy, LASTOTAL, inWLANs. A size-based scheduling policy is a priority policy where the priority of a flow is based on its size. LASTOTAL replaces the legacy IP level FIFO scheduler at the access point. The lower protocol layers, and especially the MAC 802.11 layer are left unchanged. We demonstrate using realistic synthetic workloads, that LAS-TOTAL solves the unfairness issue due to DCF in 802.11 WLANs and ensures small response times to the majority of the flows under any load conditions. The latter property is desirable as short flows correspond to interactive applications and maintaining low response times for those flows despite load variations, significantly improves user experience. We also introduce and validate Markovian queuing models to assess the response time of the access point for both FIFO and LASTOTAL.

The performance of peer-to-peer file replication comes from its piece and peer selection strategies. Two such strategies have been introduced by the BitTorrent protocol: the rarest first and choke algorithms. Whereas it is commonly admitted that BitTorrent performs well, recent studies have proposed the replacement of the rarest first and choke algorithms in order to improve efficiency and fairness. In this paper, we use results from real experiments to advocate that the replacement of the rarest first and choke algorithms cannot be justified in the context of peer-to-peer file replication in the Internet.We instrumented a BitTorrent client and ran experiments on real torrents with different characteristics. Our experimental evaluation is peer oriented, instead of tracker oriented, which allows us to get detailed information on all exchanged messages and protocol events. We go beyond the mere observation of the good efficiency of both algorithms. We show that the rarest first algorithm guarantees close to ideal diversity of the pieces among peers. In particular, on our experiments, replacing the rarest first algorithm with source or network coding solutions cannot be justified. We also show that the choke algorithm in its latest version fosters reciprocation and is robust to free riders. In particular, the choke algorithm is fair and its replacement with a bit level tit-for-tat solution is not appropriate. Finally, we identify new areas of improvements for efficient peer-to-peer file replication protocols.

In this paper we adopt a simulation approach to study the performance of the BitTorrent protocol in terms of the entropy that qualifies a torrent and the structure of the overlay used to distribute the content. We find that the entropy of a torrent, defined as the diversity that characterizes the distribution of pieces of the content, plays an important role for the system to achieve optimal performance. We then relate the performance of BitTorrent with the characteristics of the distribution overlay built by the peers taking part in the torrent. Our results show that the number of connections a given peer maintains with other peers and the fraction of those connections initiated by the peer itself are key factors to sustain a high entropy, hence an optimal system performance. Those results were obtained for a realistic choice of torrent sizes and system parameters, under the assumption of a flash-crowd peer arrival pattern.

While the applications using the Internet have changed over time, TCP is still the dominating transport protocol that carries over 90% of the total traffic. Throughput is the key performance metric for long TCP connections. The achieved throughput results from the aggregate effects of the network path, the parameters of the TCP end points, and the application on top of TCP. Finding out which of these factors is limiting the throughput of a TCP connection -- referred to as TCP root cause analysis -- is important for end users that want to understand the origins of their problems, ISPs that need to troubleshoot their network, and application designers that need to know how to interpret the performance of the application. In this paper, we revisit TCP root cause analysis by first demonstrating the weaknesses of a previously proposed flight-based approach. We next discuss in detail the different possible limitations and highlight the need to account for the application behavior during the analysis process. The main contribution of this paper is a new approach based on the analysis of time series extracted from packet traces. These time series allow for a quantitative assessment of the different causes with respect to the resulting throughput. We demonstrate the interest of our approach on a large BitTorrent dataset.

Peer-to-peer distributed hash table (DHT) systems make it simple to discover specific data when their complete identifiers - or keys - are known in advance. In practice, however, users looking up resources stored in peer-to-peer systems often have only partial information for identifying these resources. We describe techniques for indexing data stored in peer-to-peer DHT networks, and discovering the resources that match a given user query. Our system creates multiple indexes, organized hierarchically, which permit users to locate data even using scarce information, although at the price of a higher lookup cost. The data itself is stored on only one (or few) of the nodes. Experimental evaluation demonstrates the effectiveness of our indexing techniques on a distributed peer-to-peer bibliographic database with realistic user query workloads.

The Least Attained Service (LAS) scheduling policy, when used for scheduling packets over the bottleneck link of an Internet path, can greatly reduce the average flow time for short flows while not significantly increasing the average flow time for the long flows that share the same bottleneck. No modification of the packet headers is required to implement the simple LAS policy. However, previous work has also shown that a drawback of the LAS scheduler is that, when link utilization is greater than 70%, long flows experience large jitter in their packet transfer times as compared to the conventional First-Come-First-Serve (FCFS) link scheduling. This paper proposes and evaluates new differentiated LAS scheduling policies that reduce the jitter for long flows that are identified as "priority" flows.To evaluate the new policies, we develop analytic models to estimate average flow transfer time as a function of flow size, and average packet transmission time as a function of position in the flow, for the single-bottleneck "dumbbell topology" used in many ns simulation studies. Models are developed for FCFS scheduling, LAS scheduling, and each of the new differentiated LAS scheduling policies at the bottleneck link. Over a wide range of configu-rations, the analytic estimates agree very closely with the ns estimates. Thus, the analytic models can be used instead of simulation for comparing the policies with respect to mean flow transfer time (as a function of flow size) and mean packet transfer time. Furthermore, an initial discrepancy between the analytic and simulation estimates revealed errors in the parameter values that are often specified in the widely used ns Web workload generator. We develop an improved Web workload specification, which is used to estimate the packet jitter for long flows (more accurately than with previous simulation workloads).Results for the scheduling policies show that a particular policy, LAS-log, greatly improves the mean flow transfer time for priority long flows while providing performance similar to LAS for the ordinary flows. Simulations show that the LAS-log policy also greatly reduces the jitter in packet delivery times for the priority flows.

Recent studies of Internet traffic have shown that flow size distributions often exhibit a high variability property in the sense that most of the flows are short and more than half of the total load is constituted by a small percentage of the largest flows. In the light of this observation, it is interesting to revisit scheduling policies that are known to favor small jobs in order to quantify the benefit for small and the penalty for large jobs. Among all scheduling policies that do not require knowledge of job size, the least attained service (LAS) scheduling policy is known to favor small jobs the most. We investigate the M/G/1/LAS queue for both, load ? < 1 and ? = 1. Our analysis shows that for job size distributions with a high variability property, LAS favors short jobs with a negligible penalty to the few largest jobs, and that LAS achieves a mean response time over all jobs that is close to the mean response time achieved by SRPT.Finally, we implement LAS in the ns-2 network simulator to study its performance benefits for TCP flows. When LAS is used to schedule packets over the bottleneck link, more than 99% of the shortest flows experience smaller mean response times under LAS than under FIFO and only the largest jobs observe a negligible increase in response time. The benefit of using LAS as compared to FIFO is most pronounced at high load.

The need to guarantee quality-of-service (QoS) to multimedia applications leads to a tight integration between the routing and forwarding functions in the Internet. multiprotocol label switching tries to provide a global solution for this integration. In this context, multipoint-to-point (m2p) networks appear as a key architecture since they provide a cheaper way to connect edge nodes than point-to-point connections. M2p networks have been mainly studied for their load balancing ability. In this paper, we go a step further: we propose and evaluate a traffic management scheme that provides deterministic QoS guarantees for multimedia sources in an m2p network. We first derive an accurate upper bound on the end-to-end delay in an m2p architecture based on the concept of additivity. Broadly speaking, an m2p network is additive if the maximum end-to-end delay is equal to the sum of local maximum delays. We then introduce two admission control algorithms for additive networks: a centralized algorithm and a distributed algorithm and discuss their complexity and their scalability.