Taming the elephants: New TCP slow start
Injong Rhee2011, Computer Networks
Sign up for access to the world's latest research
checkGet notified about relevant papers
checkSave papers to use in your research
checkJoin the discussion with peers
checkTrack your impact
Abstract
Standard slow start does not work well under large bandwidthdelay product (BDP) networks. We find two causes of this problem in existing three popular operating systems, Linux, FreeBSD and Windows XP. The first cause is that because of the exponential increase of cwnd during standard slow start, heavy packet losses occur. Recovering from heavy packet losses puts extremely high load on end systems which renders the end systems completely unresponsive for a long time, resulting in a long blackout period of no transmission. This problem commonly occurs with the three operating systems. The second cause is that some of proprietary protocol optimizations applied for slow start by these operating systems to relieve the system load happen to slow down the loss recovery followed by slow start. To remedy this problem, we propose a new slow start algorithm, called Hybrid Start (HyStart) that finds a "safe" exit point of slow start at which slow start can finish and safely move to congestion avoidance without causing any heavy packet losses. HyStart uses ACK trains and RTT delay samples to detect whether (1) the forward path is congested or (2) the current size of congestion window has reached the available capacity of the forward path. HyStart is a plug-in to the TCP sender and does not require any change in TCP receivers. We implemented HyStart for TCP-NewReno and TCP-SACK in Linux and compare its performance with five different slow start schemes with the TCP receivers of the three different operating systems in the Internet and also in the lab testbeds. Our results indicate that HyStart works consistently well under diverse network environments including asymmetric links and high and low BDP networks. Especially with different operating system receivers (Windows XP and FreeBSD), HyStart improves the start-up throughput of TCP more than 2 to 3 times.
Related papers
Improving TCP Start-Up Behavior In High-Speed Networks
Workshop on High- …, 2003
Improving TCP Startup Performance using Active Measurements: Algorithm and Evaluation
TCP Slow Start exponentially increases the congestion window size to detect the proper congestion window for a network path. This often results in significant packet loss, while breaking off Slow Start using a limited slow start threshold may lead to an overly conservative congestion window size. This problem is especially severe in high speed networks. In this paper we present a new TCP startup algorithm, called Paced Start, that incorporates an available bandwidth probing technique into the TCP startup algorithm. Paced Start is based on the observation that when we view the TCP startup sequence as a sequence of packet trains, the difference between the data packet spacing and the acknowledgement spacing can yield valuable information about the available bandwidth. Slow Start ignores this information, while Paced Start uses it to quickly estimate the proper congestion window for the path. For most flows, Paced Start transitions into congestion avoidance mode faster than Slow Start, has a significantly lower packet loss rate, and avoids the timeout that is often associated with Slow Start. This paper describes the Paced Start algorithm and uses simulation and real system experiments to characterize its properties.
CapStart: An Adaptive TCP Slow Start for High Speed Networks
2009 First International Conference on Evolving Internet, 2009
In this document, we introduce Cap Start TCP, an adaptive Slow Start scheme that consistently achieves fast TCP file transfer times regardless of high speed network scenario. Once the TCP session is established, we estimate TCP session path capacity scenario, and tune the transport protocol to deliver fast transfer times. We demonstrate significant transaction performance improvements, of as much as three times faster completion times in transcontinental high speed network experiments for various capacity scenarios.
AFStart: An adaptive fast TCP slow start for wide area networks
2012 IEEE International Conference on Communications (ICC), 2012
Transmission Control Protocol (TCP) slow start degrades TCP performance under conditions of long-distance and high end-to-end latency, i.e., inherent characteristics of wide area networks (WANs). In this paper, we propose a new TCP slow start algorithm for WANs, called Adaptive Fast Start (AFStart), which incorporates an inline available bandwidth measurement over TCP technique into TCP slow start to set the slow start threshold adaptively and adjusts the congestion window intelligently. The performance of AFStart is evaluated through simulations using the dumb-bell topology and parking-lot topology by applying AFStart to Fast TCP. The simulation results show that AFStart can ramp up the congestion window from its initial value to the slow start threshold more quickly and smoothly than standard slow start, and AFStart achieves higher network link utilization and TCP throughput during the slow start than Fast TCP.
TCP performance
Communications Magazine, IEEE, 2001
he Transmission Control Protocol (TCP) and the User Datagram Protocol (UDP) are both IP transport-layer protocols. UDP is a lightweight protocol that allows applications to make direct use of the unreliable datagram service provided by the underlying IP service. UDP is commonly used to support applications that use simple query/response transactions, or applications that support real-time communications. TCP provides a reliable data-transfer service, and is used for both bulk data transfer and interactive data applications. TCP is the major transport protocol in use in most IP networks, and supports the transfer of over 90 percent of all traffic across the public Internet today. Given this major role for TCP, the performance of this protocol forms a significant part of the total picture of service performance for IP networks. In this article we examine TCP in further detail, looking at what makes a TCP session perform reliably and well. This article draws on material published in the Internet Performance Survival Guide [1]. Overview of TCP TCP is the embodiment of reliable end-to-end transmission functionality in the overall Internet architecture. All the functionality required to take a simple base of IP datagram delivery and build upon this a control model that implements reliability, sequencing, flow control, and data streaming is embedded within TCP [2] .
Simulation Analysis of a New Startup Algorithm for TCP New Reno
Standard TCP (New Reno) is vulnerable to startup effects that cause loss of connection setup packets or result in long round trip time (RTT) greater than 1-second. When either of these events occurs, TCP New Reno resets its congestion state by reducing initial congestion window (IW) and slow-start threshold (ssthresh) values to 1 and 2 maximum segment size (MSS) respectively. In this condition, TCP requires multiple round trips to complete delaysensitive transactions, thus resulting in poor userexperience. This paper presents a new congestion control algorithm that makes TCP more responsive by increasing its robustness against startup losses. Our main contribution in this paper is performing extensive simulation studies to investigate dynamics of the proposed algorithm. The main result obtained shows that an average latency gain of 15 RTTs can be achieved at up to 90% link utilisation, with a packet loss rate (PLR) of 1%.
Modeling TCP Startup Performance*
Journal of Mathematical Sciences, 2014
The paper proposes a Markovian approach to the performance evaluation of the ESSE (early slow start exit) modification of the TCP congestion control mechanism. ESSE takes advantage of estimating the optimal pipesize at the sender side to properly select initial slow start threshold. Previous simulative experiments have shown that ESSE allows one to speed-up TCP connections and significantly reduces the packet drop rate at the bottleneck. This work takes a step further in understanding the ESSE behavior by developing a model of TCP source to evaluate the influence of different settings of slow start threshold on a TCP performance. As confirmed by comparison with simulations, the model provides, significantly faster than simulations, accurate estimates of typical performance indicators such as the average completion time and average drop rate of short-lived TCP connections.
Delay-based congestion avoidance for TCP
IEEE/ACM Transactions on Networking, 2003
The set of TCP congestion control algorithms associated with TCP/Reno (e.g., slow-start and congestion avoidance) have been crucial to ensuring the stability of the Internet. Algorithms such as TCP/NewReno (which has been deployed) and TCP/Vegas (which has not been deployed) represent incrementally deployable enhancements to TCP as they have been shown to improve a TCP connection's throughput without degrading performance to competing flows. Our research focuses on delay-based congestion avoidance algorithms (DCA), like TCP/Vegas, which attempt to utilize the congestion information contained in packet round-trip time (RTT) samples. Through measurement and simulation, we show evidence suggesting that a single deployment of DCA (i.e., a TCP connection enhanced with a DCA algorithm) is not a viable enhancement to TCP over high-speed paths. We define several performance metrics that quantify the level of correlation between packet loss and RTT. Based on our measurement analysis we find that although there is useful congestion information contained within RTT samples, the level of correlation between an increase in RTT and packet loss is not strong enough to allow a TCP/Sender to reliably improve throughput. While DCA is able to reduce the packet loss rate experienced by a connection, in its attempts to avoid packet loss, the algorithm will react unnecessarily to RTT variation that is not associated with packet loss. The result is degraded throughput as compared to a similar flow that does not support DCA.
B-TCP: a new startup strategy for speeding up mice flows
2008
This paper investigates the degradation of to mice flows caused by standard TCP congestion control mechanisms, especially during the Slow Start phase what can cause multiple packet losses. Thus, a modified TCP startup mechanism was proposed. The Burst TCP (B-TCP) is an intuitive TCP startup mechanism that employs a responsive window growth scheme to improve performance of small flows. Moreover, B-TCP is easy to implement and requires TCP adjustment at the sender side only. Simulation experiments show that B-TCP can significantly reduce transfer times and packet losses of small flows without causing damage to large flows.
Related topics
Related papers
Tcp startup performance in large bandwidth networks
2004
Abstract Next generation networks with large bandwidth and long delay pose a major challenge to TCP performance, especially during the startup period.
An Enhanced Slow-Start Mechanism for TCP Vegas
11th International Conference on Parallel and Distributed Systems (ICPADS'05), 2005
In this article, we present a new slow-start variant, which improves the throughput of TCP Vegas, we call this new mechanism Gallop-Vegas which quickly ramps up to the available bandwidth and reduces the burstiness during the slow-start phase. Since TCP (Transmission Control Protocol) is known to send bursts of packets during its slowstart phase due to the fast increase of window size and the ACK-clock based transmission. This phenomenon causes TCP Vegas to change from slow-start phase to congestionavoidance phase too early in the large BDP (bandwidthdelay product) links. Therefore, in Gallop-Vegas, we increase the congestion window size with a rate between exponential growth and linear growth during slow-start phase. Our extensive simulation results show that Gallop-Vegas significantly improves the performance during the slow-start phase. Furthermore, it is implementation feasible because only sending part needs to be modified.
A Sender-Side TCP Enhancement for Startup Performance in High-Speed Long-Delay Networks
Many previous studies have shown that traditional TCP slow-start algorithm suffers performance degradation in high-speed and long-delay networks. This paper presents a sender-side enhancement, which makes use of TCP Vegas congestion-detecting scheme to monitor the router queue, and accordingly refines slow-start window evolution by introducing a two-phase approach to probe bandwidth more efficiently. Moreover, it achieves good fairness of bandwidth utilization in coexistence of multiple connections. Simulation results show that, compared with traditional slow-start and many other enhancements, it is able to significantly improve the startup performance without adversely affecting coexisting TCP connections.
TCP slow start, congestion avoidance, fast retransmit, and fast recovery algorithms
2001
This document specifies an Internet standards track protocol for the Internet community, and requests discussion and suggestions for improvements. Please refer to the current edition of the "Internet Official Protocol Standards" (STD 1) for the standardization state and status of this protocol. Distribution of this memo is unlimited.
Modeling a New Startup Algorithm for TCP New Reno
Standard TCP (New Reno) is vulnerable to startup effects that cause loss of connection setup packets or result in long round trip time (RTT) greater than 1-second. When either of these events occurs, TCP New Reno resets its congestion state by reducing initial congestion window (IW) and slow-start threshold (ssthresh) values to 1 and 2 maximum segment size (MSS) respectively. In this condition, TCP requires multiple round trips to complete delaysensitive transactions, thus resulting in poor userexperience. This paper presents a new congestion control algorithm that makes TCP more responsive by increasing its robustness against startup losses. Our main contribution in this paper is derivation of a stochastic model that yields a simple expression for computing the latency of a short-lived transaction as a function of IW, ssthresh and bandwidthdelay product (BDP) of an uncongested link.
Hybrid slow start for high-bandwidth and long-distance networks
2008
Slow Start is a technique to probe for unknown and time-varying available bandwidth of a network path. A sender increases its congestion window by one for each ACK received (when ACKs are not delayed), which effectively doubles its congestion window when receiving ACKs for all the packets within a congestion window. Even if an exponential increase of congestion window during Slow Start grabs unused bandwidth quite well, a large number of packet losses within an RTT is inevitable because of its overshooting. Furthermore, for fast and long distance networks, a large number of packet losses would result in unnecessarily long timeouts and create system performance bottlenecks related to handling the recovery of lost packets. We propose a new algorithm, called Hybrid Slow Start that maintains the existing Slow Start mechanism of TCP-NewReno but provides trustworthy signals to Slow Start for safely switching to Congestion Avoidance without incurring an extremely large number of packet losses. Hybrid Slow Start uses two pieces of information-ACK train length and increase in packet delays. By measuring ACK train length, a TCP sender roughly infers the maximum number of packets in flight which is typically smaller than the Bandwidth Delay Product (BDP) of the path if taken into account cross traffic and routing delays along the path. Increase in delays for the first few packets in each RTT round during Slow Start strongly indicates the path is getting congested by other traffic. Hybrid Slow Start is easy to implement using a only very small set of TCP state variables available in standard TCP. We validate our claims by applying Hybrid Slow Start to CUBIC and testing it under a realistic mix of background traffic with TCP receivers implemented in FreeBSD, Windows and Linux.
Gallop-Vegas: An enhanced slow-start mechanism for TCP Vegas
2006
In this article, we present a new slow-start variant, which improves the throughput of transmission control protocol (TCP) Vegas. We call this new mechanism Gallop-Vegas because it quickly ramps up to the available bandwidth and reduces the burstiness during the slow-start phase. TCP is known to send bursts of packets during its slow-start phase due to the fast window increase and the ACK-clock based transmission. This phenomenon causes TCP Vegas to change from slow-start phase to congestionavoidance phase too early in the large bandwidth-delay product (BDP) links. Therefore, in Gallop-Vegas, we increase the congestion window size with a rate between exponential growth and linear growth during slow-start phase. Our analysis, simulation results, and measurements on the Internet show that Gallop-Vegas significantly improves the performance of a connection, especially during the slow-start phase. Furthermore, it is implementation feasible because only sending part needs to be modified.
Comparison of methods for accelerating the response time of TCP sessions over long delay paths
2013 IEEE Symposium on Computers and Communications (ISCC), 2013
Many modern Internet applications require fast client-server interaction to maximise user interactivity. This need has caused new work to be proposed to working groups of the Internet Engineering Task Force (IETF), which seek to remove bottlenecks when starting sessions. RFC 6298 improves Transmission Control Protocol (TCP) responsiveness by retransmitting lost SYNs, or first data segments sent after the three-way handshake (3WHS), with a reduced initial retransmission timeout (RTO) of 1 second from the previous value of 3 seconds. This benefits a large number of Internet connections. However, there is growing use of wireless and other long-haul radio technologies that yield round trip times (RTT) greater than 1 second. This means a significant number of connections will suffer spurious retransmissions with associated performance penalties. This paper examines whether RFC 6298 can cooperate with a set of proposed methods to extend performance benefits to long delay paths that may experience startup loss. It was found that a slight modification of TCP congestion-state initialization after the 3WHS significantly improves performance, especially if combined with SYN duplication and an increased Initial Window (IW). We further explain how a deprecated proposal to seed the RTO based on the SYN round trip time (RTT) can be updated to become suited for general deployment. Proposed modifications are evaluated, by analysis and using ns2 simulations, showing improved overall responsiveness for short delay paths as well as networks paths with shared bandwidth and appreciable delay (e.g. wireless/satellite networks).
Evaluating Quick-Start for TCP
2005
This paper explores the Quick-Start mechanism, designed to allow transport protocols to explicitly request permission from the routers along a network path to send at a higher rate than normally allowed by traditional congestion control mechanisms. If the routers are underutilized, they may approve the sender's request for a higher sending rate; otherwise the sender uses the default congestion control algorithms. This paper discusses some of the design issues of Quick-Start, and evaluates the potential benefits, costs and implications of Quick-Start in different networking environments. Using simulations, we evaluate several different algorithms that routers could use to process a Quick-Start Request. This evaluation explores tradeoffs between the fraction of Quick-Start requests that are approved and the fraction of approved Quick-Start requests that result in increasing network congestion. In addition, the paper discusses the security implications of using Quick-Start and some possible mitigations for the vulnerabilities.
Restricted Slow-Start for TCP
2005 IEEE International Conference on Cluster Computing, 2005
In network protocol research a common goal is optimal bandwidth utilization, while still being network friendly. The drawback of TCP in networks with large bandwidth-delay products due to its AIMD based congestion control mechanism is well known. The congestion control algorithm of TCP has two phases namely slow-start phase and congestion-avoidance phase. Many researchers have focused on modifying the congestion avoidance phase of the algorithm. In this work, we propose a modification to the slow-start phase of the algorithm to achieve better performance. Restricted slow-start algorithm is a simple sender side alteration to the TCP congestion window update algorithm.
Loading Preview
Sorry, preview is currently unavailable. You can download the paper by clicking the button above.