Search

TCP does not work well in modern cellular networks because the current congestion-window-based (cwnd-based) congestion control mechanism intimately couples congestion control and packet dispatch, which provides TCP with only indirect control of the effective data rate. The throughput degradation arising from the cwnd-based mechanism is especially serious when the uplink is congested. We describe PropRate, a new rate-based TCP algorithm that directly regulates the packets in the bottleneckbuffer to achieve a trade-off in terms of delay and throughput along a more efficient frontier than conventional cwnd-based TCP variants. To the best of our knowledge, PropRate is the first TCP algorithm that allows an application to set and achieve a target average latency, if the network conditions allow for it. Also, unlike the cwnd-based TCP mechanism, our new rate-based TCP mechanism is significantly more resilient to saturated uplinks in cellular networks. PropRate does not require modifications at the receiver and is amenable to practical deployment in the base stations and proxies in mobile cellular networks.

Multi-tenant datacenters are successful because tenants can seamlessly port their applications and services to the cloud. Virtual Machine (VM) technology plays an integral role in this success by enabling a diverse set of software to be run on a unified underlying framework. This flexibility, however, comes at the cost of dealing with out-dated, inefficient, or misconfigured TCP stacks implemented in the VMs. This paper investigates if administrators can take control of a VM's TCP congestion control algorithm without making changes to the VM or network hardware. We propose AC/DC TCP, a scheme that exerts fine-grained control over arbitrary tenant TCP stacks by enforcing per-flow congestion control in the virtual switch (vSwitch). Our scheme is light-weight, flexible, scalable and can police non-conforming flows. In our evaluation the computational overhead of AC/DC TCP is less than one percentage point and we show implementing an administrator-defined congestion control algorithm in the vSwitch (i.e., DCTCP) closely tracks its native performance, regardless of the VM's TCP stack.

The popularity of smartphones and smartphone applications means that data is the dominant traffic type in current mobile networks. In this paper we present our work on a systematic investigation into facets of the LTE/EPC architecture that impact the performance of TCP as the predominant transport layer protocol used by applications on mobile networks. We found that (1) load increase in a cell causes dramatic bandwidth reduction on UEs and significantly degrades TCP performance, (2) seamless handover causes significant TCP losses while lossless handover increases TCP segments' delay.

Reliable social connectivity and transmission of data for popular nodes are vital in multihop Ad-hoc Social Networks (ASNETs). However, congestion may occur and hence popular nodes might not achieve required bandwidth when multiple senders share data with a single receiver. The traditional Transport Control Protocol (TCP) might not be able to perform efficiently in ASNETs. Therefore, we propose a Reliable TCP for Popular Data in Socially-aware Ad-hoc Networks called RTPS. RTPS employs a popularity level approach for every single node which partitions the bandwidth among users. The reliability of popular data is ensured since bandwidth will firstly assigned to the popular sender. Preliminary results show that by delaying the acknowledgment and improving the performance of TCP, RTPS reduces collision loss in multihop ASNETs.

Many scientific disciplines rely on "Experimental Design" to study various types of systems. Experimental design refers to a planned approach to experimentation that tries to provide statistical evidence to the outcome of experiments. The networking community rarely relies on such approaches, especially for real protocol implementations. Many improvements to protocols like TCP, including the recently proposed Multipath TCP, have been evaluated by considering a relatively limited set of simulations or experiments. Multipath TCP increases the goodput of a data transfer by simultaneously using multiple interfaces. It also improves load balancing thanks to dedicated congestion control. By applying experimental design, we conduct a large set of measurements inside Mininet with the Linux kernel Multipath TCP implementation, to measure its bandwidth aggregation and load balancing. Thanks to the experimental design approach, we are able to highlight several limitations of this implementation. We identify heuristics that lead to lower than expected performance and propose improvements.

With lower latency and higher bandwidth than its predecessor 3G networks, the latest cellular technology 4G LTE has been attracting many new users. However, the interactions among applications, network transport protocol, and the radio layer still remain unexplored. In this work, we conduct an in-depth study of these interactions and their impact on performance, using a combination of active and passive measurements. We observed that LTE has significantly shorter state promotion delays and lower RTTs than those of 3G networks. We discovered various inefficiencies in TCP over LTE such as undesired slow start. We further developed a novel and lightweight passive bandwidth estimation technique for LTE networks. Using this tool, we discovered that many TCP connections significantly under-utilize the available bandwidth. On average, the actually used bandwidth is less than 50% of the available bandwidth. This causes data downloads to be longer, and incur additional energy overhead. We found that the under-utilization can be caused by both application behavior and TCP parameter setting. We found that 52.6% of all downlink TCP flows have been throttled by limited TCP receive window, and that data transfer patterns for some popular applications are both energy and network unfriendly. All these findings highlight the need to develop transport protocol mechanisms and applications that are more LTE-friendly.

TCP is designed to operate in a wide range of networks. Without any knowledge of the underlying network and traffic characteristics, TCP is doomed to continuously increase and decrease its congestion window size to embrace changes in network or traffic. In light of emerging popularity of centrally controlled Software-Defined Networks (SDNs), one might wonder whether we can take advantage of the global network view available at the controller to make faster and more accurate congestion control decisions. In this paper, we identify the need and the underlying requirements for a congestion control adaptation mechanism. To this end, we propose OpenTCP as a TCP adaptation framework that works in SDNs. OpenTCP allows network operators to define rules for tuning TCP as a function of network and traffic conditions. We also present a preliminary implementation of OpenTCP in a ~4000 node data center.

With the rapid growth of data centers, minimizing the queueing delay at network switches has been one of the key challenges. In this work, we analyze the shortcomings of the current TCP algorithm when used in data center networks, and we propose to use latency-based congestion detection and rate-based transfer to achieve ultra-low queueing delay in data centers.

We've known for a while that the Internet has ossified as a result of the race to optimize existing applications or enhance security. NATs, performance-enhancing-proxies,firewalls and traffic normalizers are only a few of the middleboxes that are deployed in the network and look beyond the IP header to do their job. IP itself can't be extended because "IP options are not an option". Is the same true for TCP?In this paper we develop a measurement methodology for evaluating middlebox behavior relating to TCP extensions and present the results of measurements conducted from multiple vantage points. The short answer is that we can still extend TCP, but extensions' design is very constrained as it needs to take into account prevalent middlebox behaviors. For instance, absolute sequence numbers cannot be embedded in options, as middleboxes can rewrite ISN and preserve undefined options. Sequence numbering also must be consistent for a TCP connection, because many middleboxes only allow through contiguous flows.We used these findings to analyze three proposed extensions to TCP. We find that MPTCP is likely to work correctly in the Internet or fallback to regular TCP. TcpCrypt seems ready to be deployed, however it is fragile if resegmentation does happen---for instance with hardware offload. Finally, TCP extended options in its current form is not safe to deploy.

Many network links in developing regions operate in the sub-packet regime, an environment where the typical per-flow throughput is less than 1 packet per round-trip time. TCP and other common congestion control protocols break down in the sub-packet regime, resulting in severe unfairness, high packet loss rates, and flow silences due to repetitive timeouts. To understand TCP's behavior in this regime, we propose a model particularly tailored to high packet loss-rates and relatively small congestion window sizes. We validate the model under a variety of network conditions.

Transmission Control Protocol (TCP) is the most popular transport layer protocol used in the current internet. However, the end-to-end throughput of TCP degrades significantly with increase in number of hops in multi-hop wireless networks. Improving the performance of TCP from frequent retransmission timeouts, which is one of the well-known problems of the end-to-end throughput degradation of TCP over multi-hop wireless networks. The loss of newly sent packets during fast recovery cannot be recovered by fast retransmission, if the sender does not have enough duplicate acknowledgments and this may lead to retransmission timeouts. In this paper, we propose a new TCP mechanism, called R-RTO (Reduction of Retransmission Timeouts) having the features of TCP NewReno, which is capable of recovering packets without waiting for retransmission timeouts and thereby increase the throughput of TCP. Our proposed mechanism consists of two schemes, called Rapid Retransmission (RR) and Estimation of Available Bandwidth (EAB). Through extensive simulations using qualnet 4.5, the results show that our mechanism achieves 10--30% improvement than key existing TCP variants.

Recently new data center topologies have been proposed that offer higher aggregate bandwidth and location independence by creating multiple paths in the core of the network. To effectively use this bandwidth requires ensuring different flows take different paths, which poses a challenge.Plainly put, there is a mismatch between single-path transport and the multitude of available network paths. We propose a natural evolution of data center transport from TCP to multipath TCP. We show that multipath TCP can effectively and seamlessly use available bandwidth, providing improved throughput and better fairness in these new topologies when compared to single path TCP and randomized flow-level load balancing. We also show that multipath TCP outperforms laggy centralized flow scheduling without needing centralized control or additional infrastructure.

Congestion control algorithms seek to optimally utilize network resources by allocating a certain rate for each user. However, malicious clients can disregard the congestion control algorithms implemented at the clients and induce congestion at bottleneck links. Thus, in an adversarial environment, the network must enforce the congestion control algorithm in order to attain the optimal network utilization offered by the algorithm. Prior work protects only a single link incident on the enforcement routers neglecting damage inflicted upon other downstream links. We present CRAFT, a capability-based scheme to secure all downstream links of a deploying router. Our goal is to enforce a network-wide congestion control algorithm on all flows. As a reference design, we develop techniques to enforce the TCP congestion control. Our design regulates all flows to share bandwidth resources in a TCP-fair manner by emulating the TCP state machine in a CRAFT router. As a result, once a flow passes a single CRAFT router, it is TCP-fair on all downstream links of that router.

Increasingly, multimedia applications need higher bandwidth to provide better quality, for example in multi-party HD video conferencing. This demanding class of interactive applications simultaneously require high bandwidth and low end-to-end latency, a conflicting combination that is poorly supported in existing transports. Conventional wisdom dictates that network applications have a choice of transport protocols between TCP, if a reliable service model is desired, or UDP, if control over timing is required. In this paper we present Paceline, an enhanced transport we have devised to support interactive, high-bandwidth applications. Paceline enhances the transport service model to support application adaptation, through prioritization to provide timely delivery of important data, and cancellation to adapt the application rate to match available bandwidth. However, contrary to conventional wisdom, Paceline has not been implemented over UDP, nor does Paceline propose changes to TCP. We believe that the deployment obstacles and duplication of effort faced by solutions that alter or replace TCP entirely outweigh the challenges of mitigating its impairments. Instead, Paceline employs several mechanisms to support timely priority order delivery and cancellation above TCP: an application-level rate controller to reduce queueing delay due to excessive socket buffering, failover among connections to handle extreme cases of congestion, and message fragmentation to reduce the granularity of preemption. Our evaluation shows that Paceline improves upon conventional end-to-end latency shortcomings of using TCP, by factor of 3 in median latency and a factor of 4 in worst case latency. Meanwhile, Paceline is able to preserve TCP's performance in terms of fairness and utilization. Finally, we compare application performance with Paceline to a representative TCP alternative, Structured Stream Transport (SST), showing Paceline to be highly competitive.

Wireless channel contention/interference problems combined with the constraints imposed by the TCP congestion control scheme have a severe impact on TCP throughput performance and fairness in a wireless mesh network. In this paper, we propose a TCP-sleek spatial TDMA scheduling algorithm to construct a sleek data and ACK packet flowing environment for TCP flows along the wireless paths. The proposed scheme considers the TCP demands of individual directional links during channel access scheduling and time slot assignment to enable the TCP congestion control mechanisms to perform as effectively as they do in wired networks, irrespective of their spatial locations and the number of hops traversed. In addition, employing link-layer retransmission can effectively hide local transmission errors from the end-to-end TCP congestion control mechanisms. The simulation results demonstrate that, under the proposed scheme, the problem of unfair bandwidth sharing to long-hop TCP flows caused by wireless communication is successfully eliminated, and the throughput disparity between long- and short-hop TCP flows is also substantially reduced. The scheme also achieves high overall network throughput.

In this paper, we present a TCP-friendly congestion control that is at the same time media-friendly (i.e. sensitive to media characteristics). These types of rate controls are more suitable for multimedia streaming applications than the classical TCP congestion control and smooth TCP-friendly rate controls. We build our media-friendly rate control by combining the notion of TCP-friendliness with a general optimization framework for bandwidth sharing in computer networks. The resulting control is sensitive to network conditions (i.e. TCP-friendly) and also to multimedia stream characteristics (i.e. media-friendly) which makes it more suitable for controlling the transmission rate of multimedia streaming applications and should improve the perceived quality of video streams in best-effort network conditions.

TCP has traditionally been considered unfriendly for real-time applications. Nonetheless, popular applications such as Skype use TCP since UDP packets cannot pass through many NATs and firewalls. Motivated by this observation, we study the delay performance of TCP for real-time media flows. We develop an analytical performance model for the delay of TCP. We use extensive experiments to validate the model and to evaluate the impact of various TCP mechanisms on its delay performance. Based on our results, we derive the working region for VoIP and live video streaming applications and provide guidelines for delay-friendly TCP settings. Our research indicates that simple application-level schemes, such as packet splitting and parallel connections, can reduce the delay of real-time TCP flows by as much as 30% and 90%, respectively.

In this paper, we propose a scheme that dynamically adjusts the slow start threshold (ssthresh) of TCP. The ssthresh estimation is used to set an appropriate ssthresh. A good ssthresh would improve the transmission performance of TCP. For the congestion avoidance state, we present a mechanism that probes the available bandwidth. We adjust the congestion window size (cwnd) appropriately by observing the round trip time (RTT) and reset the ssthresh after quick retransmission or timeout using the ssthresh estimation. Then the TCP sender can enhance its performance by using the ssthresh estimation and the observed RTT. Our scheme defines what is considered an efficient transmission rate. It achieves better utilization than other TCP versions. Simulation results show that our scheme effectively improves TCP performance. For example, when the average bottleneck bandwidth is close to 30% of the whole network bandwidth, our scheme improves TCP performance by at least 10%.

The growing popularity of the 802.11 standard for building local wireless networks has generated an extensive literature on the performance modelling of its MAC protocol. However, most of the available studies focus on the throughput analysis in saturation conditions, while very little has been done on investigating the interactions between the 802.11 MAC protocol and closed-loop transport protocols such as TCP. This paper addresses this issue by developing an analytical model to compute the stationary probability distribution of the number of backlogged nodes in a WLAN in the presence of persistent TCP-controlled download and upload data transfers, and embedding the network backlog distribution in the MAC protocol modelling. A large set of experiments conducted in a real network validates the model correctness for a wide range of configurations. A particular emphasis is devoted to investigate and explain the TCP fairness characteristics. Our analytical model and the supporting experimental outcomes demonstrate that using default settings for the capacity of devices' output queues provides a fair allocation of channel bandwidth to the TCP connections, independently of the number of downstream and upstream flows. Furthermore, we show that the TCP total throughput does not degrade by increasing the number of wireless stations.

Most of the TCP/IP applications running over mobile communication systems such as GPRS, UMTS or CDMA1X-EV/DO use TCP as the transport protocol. The end-to-end performance of an application depends significantly on the performance of TCP. In particular, when a mobile device moves between two cells an entire burst of packets might be lost. How well TCP recovers from this type of scenario in which losses are highly correlated will greatly influence the end-to-end performance of the upper layer application. Correlated losses are also introduced by drop-tail buffers, which is the dominant packet drop policy implemented in mobile communication systems. In this paper we develop an analytical model to analyse the performance of the TCP SACK-based loss recovery mechanism defined in RFC 3517 under correlated losses imposed by mobility and/or drop-tail buffers. The results show that the mean TCP SACK throughput is no better than TCP Reno. Our model includes the details of the loss recovery phase, which helps us to show the low probability of having a successful SACK recovery phase and its potentially long recovery times

Fast-growing Internet applications like streaming media and telephony prefer timeliness to reliability, making TCP a poor fit. Unfortunately, UDP, the natural alternative, lacks congestion control. High-bandwidth UDP applications must implement congestion control themselves-a difficult task-or risk rendering congested networks unusable. We set out to ease the safe deployment of these applications by designing a congestion-controlled unreliable transport protocol . The outcome, the Datagram Congestion Control Protocol or DCCP, adds to a UDP-like foundation the minimum mechanisms necessary to support congestion control. We thought those mechanisms would resemble TCP's, but without reliability and, especially, cumulative acknowledgements, we had to reconsider almost every aspect of TCP's design. The resulting protocol sheds light on how congestion control interacts with unreliable transport, how modern network constraints impact protocol design, and how TCP's reliable bytestream semantics intertwine with its other mechanisms, including congestion control.

Past research studying the operations of TCP over wireless/mobile ad hoc networks has shown that TCP cannot be adopted as-is for use in mobile ad hoc networks (MANET) to achieve reasonable performance. This is because the assumptions made during designing TCP were made to make TCP work for wired networks, not MANETs. Consequently, most past research evaluating the performance of MANET routing protocols has concentrated on using CBR traffic over UDP for evaluation. In this paper, we distinguish practical use of TCP from use of TCP for evaluation purposes, and examine whether using TCP as an evaluation tool to evaluate MANET routing protocols will yield any new insights, even though we readily agree with past research that using TCP as-is (that is, as packaged for wired networks) for practical use over MANET is not a good approach. We motivated this idea because even though TCP does not function well over MANET, the fundamental expectations TCP has over the network are "universal" metrics which MANET researchers should strive for as well.Using simulations, we show that by studying the performance of MANET routing protocols using TCP traffic, we obtain a different understanding of the protocols studied, and this understanding cannot be obtained via applying CBR/UDP traffic. In this spirit, we urge MANET routing protocol designers to use TCP as an evaluation tool during the protocol design process.

Predicting the throughput of large TCP transfers is important for a broad class of applications. This paper focuses on the design, empirical evaluation, and analysis of TCP throughput predictors. We first classify TCP throughput prediction techniques into two categories: Formula-Based (FB) and History-Based (HB). Within each class, we develop representative prediction algorithms, which we then evaluate empirically over the RON testbed. FB prediction relies on mathematical models that express the TCP throughput as a function of the characteristics of the underlying network path. It does not rely on previous TCP transfers in the given path, and it can be performed with non-intrusive network measurements. We show, however, that the FB method is accurate only if the TCP transfer is window-limited to the point that it does not saturate the underlying path, and explain the main causes of the prediction errors. HB techniques predict the throughput of TCP flows from a time series of previous TCP throughput measurements on the same path, when such a history is available. We show that even simple HB predictors, such as Moving Average and Holt-Winters, using a history of few and sporadic samples, can be quite accurate. On the negative side, HB predictors are highly path-dependent. We explain the cause of such path dependencies based on two key factors: the load on the path and the degree of statistical multiplexing.

TCP is widely used in commercial multimedia streaming systems, with recent measurement studies indicating that a significant fraction of Internet streaming media is currently delivered over HTTP/TCP. These observations motivate us to develop analytic performance models to systematically investigate the performance of TCP for both live and stored-media streaming. We validate our models via ns simulations and experiments conducted over the Internet. Our models provide guidelines indicating the circumstances under which TCP streaming leads to satisfactory performance, showing, for example, that TCP generally provides good streaming performance when the achievable TCP throughput is roughly twice the media bitrate, with only a few seconds of startup delay.

A number of recent research studies have explored ways to improve TCP throughput in IEEE 802.11 mobile ad hoc networks. In particular, hidden terminal effects caused by interference have been shown to degrade TCP performance and have been dealt with, in previous work, by restricting the maximum congestion window size. We have pursued an alternative approach and have developed a TCP variant which adjusts the sending rate increase to achieve competitive goodput for TCP connections. In this paper, we demonstrate that a slower sending rate increase during the congestion avoidance phase of TCP leads to improved performance for TCP Reno while eliminating the negative effects inherent in restricting the maximum congestion window size. A performance comparison against existing solutions under various mobility conditions is also included.

Theory and experiments show that as the per-flow product of bandwidth and latency increases, TCP becomes inefficient and prone to instability, regardless of the queuing scheme. This failing becomes increasingly important as the Internet evolves to incorporate very high-bandwidth optical links and more large-delay satellite links.To address this problem, we develop a novel approach to Internet congestion control that outperforms TCP in conventional environments, and remains efficient, fair, scalable, and stable as the bandwidth-delay product increases. This new eXplicit Control Protocol, XCP, generalizes the Explicit Congestion Notification proposal (ECN). In addition, XCP introduces the new concept of decoupling utilization control from fairness control. This allows a more flexible and analytically tractable protocol design and opens new avenues for service differentiation.Using a control theory framework, we model XCP and demonstrate it is stable and efficient regardless of the link capacity, the round trip delay, and the number of sources. Extensive packet-level simulations show that XCP outperforms TCP in both conventional and high bandwidth-delay environments. Further, XCP achieves fair bandwidth allocation, high utilization, small standing queue size, and near-zero packet drops, with both steady and highly varying traffic. Additionally, the new protocol does not maintain any per-flow state in routers and requires few CPU cycles per packet, which makes it implementable in high-speed routers.

Most of the traffic in today's Internet is controlled by the Transmission Control Protocol (TCP). Hence, the performance of TCP has a significant impact on the performance of the overall Internet. TCP is a complex protocol with many user-configurable parameters and a range of different implementations. In addition, research continues to produce new developments in congestion control mechanisms and TCP options, and it is useful to trace the deployment of these new mechanisms in the Internet. As a final concern, the stability and fairness of the current Internet relies on the voluntary use of congestion control mechanisms by end hosts. Therefore it is important to test TCP implementations for conformant end-to-end congestion control. Since web traffic forms the majority of the TCP traffic, TCP implementations in today's web servers are of particular interest. We have developed a tool called TCP Behavior Inference Tool (TBIT) to characterize the TCP behavior of a remote web server. In this paper, we describe TBIT, and present results about the TCP behaviors of major web servers, obtained using this tool. We also describe the use of TBIT to detect bugs and non-compliance in TCP implementations deployed in public web servers.

Our research focuses on end-to-end congestion avoidance algorithms that use round trip time (RTT) fluctuations as an indicator of the level of network congestion. The algorithms are referred to as delay-based congestion avoidance or DCA. Due to the economics associated with deploying change within an existing network, we are interested in an incrementally deployable enhancement to the TCP/Reno protocol. For instance, TCP/Vegas, a DCA algorithm, has been proposed as an incremental enhancement. Requiring relatively minor modifications to a TCP sender, TCP/Vegas has been shown to increase end-to-end TCP throughput primarily by avoiding packet loss. We study DCA in today's best effort Internet where IP switches are subject to thousands of TCP flows resulting in congestion with time scales that span orders of magnitude. Our results suggest that RTT-based congestion avoidance may not be reliably incrementally deployed in this environment. Through extensive measurement and simulation, we find that when TCP/DCA (i.e., a TCP/Reno sender that is extended with DCA) is deployed over a high speed Internet path, the flow generally experiences degraded throughput compared to an unmodified TCP/Reno flow. We show (1) that the congestion information contained in RTT samples is not sufficient to predict packet loss reliably and (2) that the congestion avoidance in response to delay increase has minimal impact on the congestion level over the path when the total DCA traffic at the bottleneck consumes less than 10% of the bottleneck bandwidth.

Mobile ad hoc networks have attracted attention lately as a means of providing continuous network connectivity to mobile computing devices regardless of physical location. Recent research has focused primarily on the routing protocols needed in such an environment. In this paper, we investigate the effects that link breakage due to mobility has on TCP performance. Through simulation, we show that TCP throughput drops significantly when nodes move, due to TCP's inability to recognize the difference between link failure and congestion. We also analyze specific examples, such as a situation where throughput is zero for a particular connection. We introduce a new metric, expected throughput, for the comparison of throughput in multi-hop networks, and then use this metric to show how the use of explicit link failure notification (ELFN) techniques can significantly improve TCP performance.

In this paper we develop a simple analytic characterization of the steady state throughput, as a function of loss rate and round trip time for a bulk transfer TCP flow, i.e., a flow with an unlimited amount of data to send. Unlike the models in [6, 7, 10], our model captures not only the behavior of TCP's fast retransmit mechanism (which is also considered in [6, 7, 10]) but also the effect of TCP's timeout mechanism on throughput. Our measurements suggest that this latter behavior is important from a modeling perspective, as almost all of our TCP traces contained more time-out events than fast retransmit events. Our measurements demonstrate that our model is able to more accurately predict TCP throughput and is accurate over a wider range of loss rates.

With the growth of high performance networking, a single host may have simultaneous connections that vary in bandwidth by as many as six orders of magnitude. We identify requirements for an automatically-tuning TCP to achieve maximum throughput across all connections simultaneously within the resource limits of the sender. Our auto-tuning TCP implementation makes use of several existing technologies and adds dynamically adjusting socket buffers to achieve maximum transfer rates on each connection without manual configuration.Our implementation involved slight modifications to a BSD-based socket interface and TCP stack. With these modifications, we achieved drastic improvements in performance over large bandwidth delay paths compared to the default system configuration, and significant reductions in memory usage compared to hand-tuned connections, allowing servers to support at least twice as many simultaneous connections.

We describe tcpanaly, a tool for automatically analyzing a TCP implementation's behavior by inspecting packet traces of the TCP's activity. Doing so requires surmounting a number of hurdles, including detecting packet filter measurement errors, coping with ambiguities due to the distance between the measurement point and the TCP, and accommodating a surprisingly large range of behavior among different TCP implementations. We discuss why our efforts to develop a fully general tool failed, and detail a number of significant differences among 8 major TCP implementations, some of which, if ubiquitous, would devastate Internet performance. The most problematic TCPs were all independently written, suggesting that correct TCP implementation is fraught with difficulty. Consequently, it behooves the Internet community to develop testing programs and reference implementations.