The Round-Trip Time (Research Paper Sample)

The Round-Trip Time

Introduction
In addition to parameters such as throughput and availability for Internet quality, the response time, i.e. the round-trip time (RTT), is a crucial factor. The RTT indicates the time required to send a packet of data from a source to the receiver over a network and to transport the receiver’s response back to the transmitter. Measuring and monitoring RTTs in communication networks is important for multiple reasons, such as (a) to investigate and verify service level agreements by measuring the responsiveness of various services, (b) to allow network operators and end users to understand network behavior and performance, (c) to assist in making accurate routing and queuing decisions, and (d) to improve the achievable throughput of transport protocols, to name just a few.
The methods for measuring RTT can mainly be divided into active and passive measurement methods.
The active RTT measurement methods are based on injecting additional test packets into the measurement path. A well-known representative of this class is the ping command, which is based on the Internet Control Message Protocol (ICMP). It sends an echo request packet to a destination address that responds with an echo reply packet. The transit time can be estimated from the difference between the arrival time of the echo reply packet and the sending time of the echo request packet [1, 2]. However, there exist several significant disadvantages of active RTT measurements. ICMP is often disabled for security reasons and the additional ICMP packets lead to unwanted additional load of the network, which can falsify the measurements [3]. Active measurement techniques in general suffer from deployment flexibility limitations as all active tools need to be deployed and run on at least one of the end-hosts of each path to be measured, if not on both ends. In the case of only a couple of existing measurement paths, manual deployments and executions of such active tools could lead to temporary solutions without causing a major configuration overhead. However, in the case of reasonably large communication networks that consist of a significantly higher number of paths and hops, their probing via such manual configurations would be practically infeasible. Moreover, active tools report their final estimates in a decentralized way, either on sender or receiver hosts, which makes centralized and uniform network measurement and monitoring a more challenging task. For such a purpose, a distributed measurement architecture is needed that ideally provides the network administrators and operators with a single central host in which the measurements of any path, subpath, or hop within the overall managed network can be flexibly collected without causing any manual configuration overhead.
Passive tools, on the other hand, eliminate the disadvantages of active tools as they do not generate additional traffic, but only capture and analyze the passing real traffic at an appropriate observation point without perturbing the network traffic. They offer the important advantage of flexible placement of the monitoring software anywhere on the way between the end hosts. For these reasons, passive RTT measurement methods are more often preferred in practice [4]. For the passive measurement of RTTs, the well-known Transmission Control Protocol (TCP) is particularly suitable since a transmission of a data segment from the transmitter to the receiver requires the transport of the associated acknowledgment (ACK) segment.
Within the course of the past years, several studies have been carried out to measure RTTs in the literature. Shah et al. [5] investigated the possible relationship between the RTT and the software download time for FTP servers. Günther and Hoene [6] measured RTTs with the intention of determining the distance between WLAN nodes. Jiang and Dovrolis [7] proposed and evaluated a passive measurement methodology that estimates the distribution of RTTs for the TCP connections. The work of Strowes [2] is based on the observation of packets using the TCP header timestamp option, which is used by TCP to generate RTT measures. Aikat et al. [8] measured and analyzed the variability in RTTs within TCP connections using passive measurement techniques. Yan et al. [9] presented a novel passive measurement method that exploits the TCP timestamp option to measure path RTT at an intermediate measurement point between two end-points. Prieto et al. [4] presented a simple passive algorithm to estimate the RTT of a TCP connection in high bandwidth-delay network scenarios. Veal et al. [10] proposed two methods to passively measure and monitor changes in RTTs throughout the lifetime of a TCP connection. Jaiswall et al. [11] proposed a passive measurement methodology to infer and keep track of RTTs with a TCP connection based on the estimated value of the congestion window. Despite the plethora of studies on measuring RTT, the mutual motivation of all these studies was only focusing on measuring or analyzing the total end-to-end RTTs, disregarding the separate calculation of delays over individual hops along that path.
Differently from the rest of the studies in the literature, the aim of this study is to design, implement, and evaluate a distributed measurement architecture that not only measures the total end-to-end RTT. The delay times of the individual hops should also be measured if the connection between the client and server is established via several intermediate stations, the so-called gateway routers. The idea is to determine the cause of the delay of a packet more precisely. Particularly, the delay of a packet can have different causes, e.g., it can be caused by a congested component within the network of the respective ISP. Another scenario arises if the connection is established via two or more ISPs. In this case, it can be determined which provider along the end-to-end path has delayed the communication. Through tests in a controlled laboratory environment, the accuracy of the proposed passive RTT measurement technique has been validated with the help of active measurements. Also, the performance of the distributed architecture was experimentally evaluated to determine to what extent it is scaling, i.e. how it behaves with respect to capturing arriving frames, connections, and CPU usage as the volume of incoming network traffic is increasing.
Theoretical foundations of passive RTT measurement
End-to-end RTT measurement using SYN/ACK (SA) methodology
There are several well-known methods that can be used to measure the end-to-end RTTs over TCP connections. For example, the slow-start method measures the RTT during the slow-start phase of a TCP connection, while continuous measurement methods measure it over the entire duration of a connection [7]. In this study, the so-called SA methodology is used for passive RTT measurement, which measures the RTT when setting up a TCP connection. Compared to the other two measurement methods, it is easier to implement while also accurately measuring RTTs that are within limits of acceptable accuracy.
Setting up a TCP connection consists of a sequence of three segments, i.e. SYN, SYN/ACK, and ACK segments, and is thus called a three-way handshake (TWH). A monitoring software implementing the SA methodology, placed anywhere on the path between the client and server, measures the RTT by observing the three consecutive TWH segments. Both the arrival times of the initial SYN segment (i.e. active open) and the ACK segment following the SYN/ACK segment (i.e. passive open) are recorded. Subsequently, the end-to-end RTT of the connection can be determined by subtracting the recorded time values. The requirement for a correct measurement is that none of the three segments are lost and the segments are immediately sent by the client or the server [7] (see Figure 1).
The main difficulty in implementing the SA methodology is the filtering of the TWH segments of a TCP connection out of network traffic. The detection of the first two segments of the TWH can be undertaken without problems because the SYN or SYN/ACK flags only set in such types of segments. However, the ACK segment is unidentifiable by its ACK flag as there are segments in the network traffic where the ACK flag is set but they are not part of the TWH phase, e.g., the acknowledgments of data segments, or acknowledgments sent as responses to connection termination requests triggered by one of the end hosts. This problem and its solution will be discussed in Section 3.3 in more detail.
Although the SA methodology can be used to measure the total RTT for the end-to-end connection, its individual usage is not sufficient to calculate the delay times of the individual hops of a connection.
Proposed method of distributed RTT measurement: extended SA methodology
The placement of the probes on each gateway router allows the computation of the partial RTTs of a connection within a network path. Figure 2 illustrates the principle of distributed measurement through the use of probes on two gateway routers. In Figure 2, the first and second probes are referred to as alpha and beta, respectively. Using two gateway routers on an end-to-end path results in a total of three sections to be measured, namely the first hop between the client and the probe alpha, the second hop between the probes alpha and beta, and finally the third hop between the probe beta and server.
SYN SYN/ACK ACK RTT RTT (measured) Client Server Time Time Time Monitor Figure 1.Principle of the SA methodology.SYN SYN/ACK ACK RTT RTT (measured) Client Server Time Time Time Monitor Figure 1.Principle of the SA methodology.
For the RTT calculation of the first hop, probe alpha measures the arrival times of both SY...