97 On TCP-Induced Telehaptic Packet Loss and Jitter Vineet Gokhale, Jan Fesl Institute of Applied Informatics, University of South Bohemia, CZECH REPUBLIC, emails: {vgokhale, jfesl}@prf.jcu.cz Abstract: Telehaptic data communication (transmission telemanipulator with the feeling of touching the patient’s body of touch signals) is known to be extremely sensitive to [1]. Telehaptic communication finds potential applications in packet loss and jitter, the primary consequences of a wide variety of other domains as well, like telemaintenance, network congestion. Existing studies have established the and remote disaster management to name a few. Quality of Service (QoS) conditions that need to be Figure 1 depicts a typical telehaptic communication guaranteed for smooth telehaptic communication. framework over a shared network. The human operator (OP), Specifically, the telehaptic communication can tolerate no using the force, audio, and video feedback from the remote more than 10% packet loss and 10 ms jitter. In this paper, environment, makes certain movements in an attempt to we conduct a detailed investigation of the impact of TCP interact with and/or manipulate a remote physical object. The cross-traffic (pre-dominant traffic on shared networks) on position and velocity signals thus generated are transmitted to telehaptic packet loss and jitter. The important the remote environment via the forward channel. The robotic contribution of our study is twofold. Firstly, we discover teleoperator (TOP) at the remote location utilizes these that even during scenarios where the long term average coordinates in order to replicate OP’s movements accurately. packet loss is comfortably below its QoS limit, the Any contact between the remote object and the TOP generates instantaneous loss can far exceed this limit. Secondly, we forces, which are transmitted back to the OP along with audio demonstrate that the probability of jitter QoS violation and video feedback on the backward channel. The presence of increases with the number of concurrent TCP sources in haptic feedback has been shown to increase the immersion into the network. These effects could potentially be harmful to the remote environment, and further improve the precision of the telehaptic activity, thereby raising serious concerns on the telehaptic activity significantly [2]. designing efficient communication frameworks for minimizing telehaptic packet loss and jitter on shared networks. Keywords: Telehaptic communication, QoS, shared network, packet loss, jitter I. INTRODUCTION Everyday activities that the humans perform are largely dependent on our sensory mechanisms that aid in learning the physical properties of any real object such as size, shape, Fig. 1. Schematic representation of a telehaptic communication weight, texture, hardness, smell, and so on. The touch framework depicting the signal exchange between the human perception forms an integral part of our sensory mechanism. operator (OP) and the robotic teleoperator (TOP). When an object is held, it exerts certain forces on the hand. The muscles and the joints of the hand capture these forces and Naturally, such highly sensitive operations necessitate they are then transmitted to the brain, generating a perception accurate replication of the OP’s movements by the TOP, and map of the object. This sensory mechanism is the fundamental also timely delivery of the feedback signals to the OP. For driving force behind the innumerous forms of seamless example, large delays in haptic feedback result in sluggish interaction between humans and the physical world. Life perception of the patient’s body, thereby (potentially) leading would be lot harder if one was to light a matchstick, drive a to a wrong action by the surgeon. Additionally, large telehaptic car, or play a game of golf without the ability to feel the jitter leads to perceiving the same remote object as having physical object. variable mass, which is absurd. Note that jitter refers to the Haptics relates to the science behind the different variation in the packet delays. High packet losses may cause mechanisms of perception of real objects through the sense of improper replication of the OP’s movements accurately and/or touch. The deep research insights in this field have led to the OP being severely deprived of the feedback signals. Both these design of elegant electro-mechanical systems that have scenarios could have catastrophic effects on the ongoing enabled us to interact and manipulate virtual as well as remote telehaptic activity. Note that the packet loss in the network is a objects through the feeling of touch. consequence of queue overflows during congestion. These Telehaptic communication – the science of coding, and effects can, at times, cause irreparable damage to the patient. subsequent transmission of haptic signals over a network – has Hence, the communication network that transfers the witnessed rapid progress over the past decade. Such telehaptic feedback plays an instrumental role in determining communication has been envisaged to redefine the way we the quality of the telehaptic interaction. interact with a remote world. For example, a surgeon could Experimental studies, such as [3], have demonstrated that perform telesurgery on a distant patient through a robotic the human perception of haptic feedback can tolerate a ACIT 2018, June 1-3, 2018, Ceske Budejovice, Czech Republic 98 maximum packet loss and jitter of not more than 10% and 10 underutilization the network resources. The TCP source ms, respectively. This means that the perception of the remote increases its data rate until it detects a packet loss (indicating environment is not hampered even if at least 90% of the congestion). In response, it reduces the data rate in order to telehaptic samples reach the OP/TOP with a jitter of no more relieve the network, and thereby achieve congestion control. than 10 ms. These telehaptic packet loss and jitter constraints Once the source detects that the network is free, it begins to that need to be satisfied for a seamless telehaptic activity are increase the data rate, and this cycle continues. As can be collectively known as Quality of Service (QoS). For a smooth observed, the TCP source relies heavily on the packet loss in telehaptic activity, the network needs to guarantee QoS- the network in order to learn the available network bandwidth. compliance at all times. In general, QoS violations lead to In fact, the working principle of TCP is itself based on deteriorated perception of the remote environment, as inducing packet loss in the network. This behavior naturally explained previously. impacts the concurrent streams in the network. In addition, the It is important to note that the work in [3] treats the packet data rate variation of TCP also introduces jitter that negatively loss as a time-average entity. In other words, the work in [3] affects the telehaptic activity. In this work, we are interested in averages packet losses over an entire telehaptic session; the studying whether these packet loss and jitter effects of TCP authors discovered that when this long term average packet have any notable impact on QoS-compliance of the telehaptic loss exceeds 10%, the users started perceiving an unacceptable stream. deterioration in the perception of the remote environment. In this paper, we intend to study the impact of multiple TCP Note that the long term average packet loss refers to the cross-traffic sources on a telehaptic stream. The objective of average of the packet loss measured over the entire duration of this investigation is to gain insights into the characteristics of the telehaptic session. It is worth noting that this work does not the instantaneous telehaptic packet loss and jitter under the consider the characteristics of the instantaneous loss while influence of coexisting TCP cross-traffic sources. The establishing 10% as the packet loss criteria for smooth contribution of our work is as follows. perception. (i) We demonstrate that in a wide range of settings, even It is important to remark that in a real world scenario the though the long term average packet loss meets the QoS perception of remote objects (potentially) depends on the criteria, the instantaneous packet loss can be much higher. instantaneous packet loss rather than the long term average (ii) We show that the peak telehaptic jitter can far exceed the loss. For example, in a few network settings the instantaneous 10 ms deadline for standard network settings, and hence is packet loss is way higher than 10% (see Figures 4 and 5) extremely prone to QoS violations. despite its long term average value being below 10%. This The remainder of the paper is organized as follows. In means that a vast majority of the telehaptic samples (up to 80% Section II, we discuss in brief a few prior works available in in our simulations) do not reach the destination. As per the the literature related to the interplay between TCP and claim in [3], this implies that even when all packets (100%) telehaptic streams. Section III describes the detailed are lost over a certain interval, the users do not feel any simulation setup that we designed for our investigation. In perceptual degradation. This is incorrect as no haptic feedback Section IV, we present the results of our experiments, and in leads to improper perception of the remote world. Therefore, Section V, we state our conclusions and mention potential the instantaneous packet loss, and not the long term average directions for future research. loss, is a more relevant performance metric from the standpoint of perception in any telehaptic communication. II. RELATED WORK A telehaptic stream on a shared network, like the Internet, Only a handful of works have attempted to study the has to contend with other cross-traffic streams that are behavior of telehaptic streams on a shared network [6, 7, 8]. concurrently being served by the network. Hence, it is crucial Although these works considered network cross-traffic in their to study the influence of the coexisting cross-traffic streams on performance evaluation, negligible attention is paid to the TCP the telehaptic stream in terms of instantaneous packet loss and streams that form a major component of the overall cross- jitter. On a shared network, the telehaptic stream is guaranteed traffic. A recent work [9] conducted a comprehensive analysis to encounter Transmission Control Protocol (TCP) traffic of the effects of a single TCP stream on the long term average since TCP amounts to over 90% of the overall traffic [4]. TCP telehaptic packet loss as well as jitter. However, this work provides a reliable data communication mode, and hence investigates ignores the instantaneous packet loss. As forms the cornerstone of a wide variety of Internet services that explained earlier, the instantaneous packet loss forms a more require reliable transfer of data, such as email, file transfer, important performance metric than the long term average web browsing, and video streaming applications like YouTube, measure. Furthermore, this analysis confines the number of and Netflix. concurrent TCP streams to one. Hence, the effect of multiple For our investigation in this paper, we consider a specific TCP streams on telehaptic loss and jitter remains unexplored flavor of TCP named TCP NewReno [5]. TCP NewReno (or in this work. any TCP source in general) is a rate-adaptive transport layer protocol that controls its transmission rate depending on the III. SIMULATION SETUP congestion level in the network. The TCP source uses packet In this section, we give a detailed description of the loss as an indicator of congestion. Based on the packet loss as experimental settings considered in our simulations. The goal detected by the source, the data rate is adapted to match the of this section is to develop an understanding of the dynamics available network bandwidth, and thereby eschew of interplay between TCP and telehaptic streams when the two ACIT 2018, June 1-3, 2018, Ceske Budejovice, Czech Republic 99 traffic types share a single bottleneck link. We carry out our We now move to studying the behavior of the instantaneous investigation using NS3 – a discrete event network simulator telehaptic packet loss for a specific value of n for which the [10]. We use the single bottleneck network topology as shown average loss meets the telehaptic packet loss criteria. For this in Figure 2. H 1 and H 2 are the OP and the TOP, respectively, purpose, we choose n = 10. Note, from Figure 3, that the long of the telehaptic communication framework shown in Figure term average loss for n = 10 is approximately 10%. From 1. [S 1 , …, S n ] and [R 1 , …, R n ] are the sets of n TCP sources Figure 4, it can be seen that the instantaneous telehaptic packet and receivers, respectively. L 1 is the bottleneck link on the loss varies rapidly between 0 and 50%. In addition to the peak forward channel. Note that the data rate variation of TCP loss measurement of 50%, it can also be seen that the packet influenced the queue occupancy at B 1 , the router at the ingress loss QoS criteria gets violated regularly. Although we report of L 1 . As mentioned earlier, the TCP sources employ the instantaneous packet loss only for n = 10, we observe NewReno congestion control scheme. For telehaptic similar behavior for other values of n as well. In short, even communication, we leverage the protocol proposed in [8]. It though the long term average packet loss meets the QoS can be shown that in presence of TCP NewReno sources, a criteria, the instantaneous loss can be significantly higher. This telehaptic source employing the protocol in [8] generates confirms our conjecture that the instantaneous packet loss packets at the rate of 250 per second. The packet scheduling at should be considered as the performance metric rather than the the network queues is based on the standard droptail long term average packet loss. mechanism. Fig. 2. Single bottleneck network topology used in our simulations. Notations: H 1 and H 2 – operator and teleoperator in telehaptic communication, respectively; [S 1 ,..., S n ] – TCP sources; [R 1 ,..., R n ] Fig. 3. Evolution of long term average telehaptic packet loss as a – TCP receivers; L 1 – bottleneck link; B 1 – router at the ingress of function of the number of TCP sources. bottleneck link. It is important to remark that even though the interval over The propagation delay of each link is set to 5 ms, and hence which the QoS violation occurs is small (a maximum of 300 the one-way propagation delay between a source and its ms), this could potentially have severe artifacts considering the corresponding receiver is 15 ms. The channel capacity of L 1 is scale of sensitivity that a telehaptic activity, like telesurgery, set to 3 Mbps. The access links to L 1 have high capacities of 5 requires. Gbps. The queue size at the B 1 is configured to 15 kB. The TCP and the telehaptic packets have sizes 578 B and 512 B, respectively, unless mentioned otherwise. For the purpose of our simulations, we consider n in the range [1, 10]. However, it is worth remarking that the observations that we make regarding the telehaptic loss and jitter hold good for higher values of n as well. All sources begin the transmissions simultaneously at t = 0. We run each simulation until t = 100 s. Throughout the simulations, we record the packet loss and jitter encountered by the telehaptic sources. Fig. 4. Instantaneous telehaptic packet loss n = 10 showing IV. RESULTS significant overshoot compared to its long term average value. In this section, we present the results of our investigation of telehaptic packet loss and jitter induced by the coexisting TCP streams. We begin by reporting the packet loss, and then move to the jitter part. In Figure 3, we report the long term average packet loss seen by the telehaptic source by varying n over the considered range. It can be seen that the long term average packet loss is an increasing function of n. However, for n < 10, the long term average packet loss complies to the QoS limit of 10%. However, we note that for higher n, the average loss exceeds the QoS limit severely. For brevity, we do not report the telehaptic packet loss in the higher n regime. Fig. 5. Variation of peak instantaneous telehaptic packet loss as a function of the number of TCP sources in the network. ACIT 2018, June 1-3, 2018, Ceske Budejovice, Czech Republic 100 Having seen the instantaneous loss, we now turn towards V. CONCLUSIONS determining the peak instantaneous telehaptic packet loss in the simulations. Figure 5 shows the variation of the peak In this paper, we conducted an extensive investigation of the packet loss in the considered range of n. It can indeed be interplay between TCP NewReno and telehaptic streams. We observed that the instantaneous packet losses are substantially demonstrated that even though the long term average higher despite the long term average packet loss complying to telehaptic packet loss satisfies the QoS criteria, the the QoS requirement. Therefore, we demonstrate through instantaneous loss can far exceed the QoS limit of 10%. experiments that any guarantees on the long term average Additionally, we showed that the telehaptic stream faces packet loss do not imply any guarantees on the peak extreme jitter QoS violations for TCP packets of standard instantaneous packet loss. This suggests that in order to ensure sizes. Hence, we conclude that it is crucial to monitor and a seamless telehaptic activity, one must design communication control the number of TCP streams, as well as the size of TCP frameworks that can provide QoS guarantees on the packets in order to achieve seamless telehaptic communication instantaneous telehaptic packet loss. on a shared network. It has been shown in the past that smaller telehaptic packets In a future version of this article, we intend to propose a (relative to TCP packets) are less susceptible to losses [9]. telehaptic communication framework that mitigates the Specifically, the experiments in [9] reveal that the telehaptic detrimental effects of TCP sources. Also, studying the effects packets of size 137 B are rarely dropped by the network queues of other variants of TCP on telehaptic stream could be another in presence of a single TCP source that transmits packets of interesting avenue for future research. size 578 B. Hence, one potential solution for mitigating the REFERENCES telehaptic losses is to minimize the packet sizes. The other plausible remedy could be to design priority queueing schemes [1] R. Anderson, and M. Spong, "Bilateral control of that can serve packets carrying crucial telehaptic data with teleoperators with time delay," IEEE Transactions on higher precedence over other cross-traffic streams. Automatic Control, vol. 34, pp. 494–501, May 1989. We now move to the peak telehaptic jitter measurements. [2] C. Basdogan, C. Ho, M. A. Srinivasan, and M. Slater, "An For the packet sizes mentioned in Section III, we notice that experimental study on the role of touch in shared virtual the peak telehaptic jitter varies in the range [2.55, 6.69] ms, environments," in ACM Transactions on Computer-Human which satisfies the QoS constraint. Since the jitter is known to Interaction (TOCHI), vol. 7, pp. 443–460, 2000. be heavily dependent on the TCP packet size, for concreteness [3] P. Dev, D. Harris, and D. Gutierrez, A. Shah, and S. Senger, in exposition, we run the simulations with TCP packets of size "End-to-end performance measurement of internet based 1042 B, which is also another standard value. In Figure 6, we plot the peak instantaneous telehaptic jitter as a function of n. medical applications," in Proceedings of the Annual It can be seen that the peak jitter is a non-decreasing function Symposium of the American Medical Informatics of n. Further, for n > 5, the jitter QoS condition is severely Association, Nov 2002. violated. This implies that higher the number of concurrent [4] S. Ryu, C. Rump, and C. Qiao, "Advances in internet TCP streams, larger is the probability of violation of the congestion control," in IEEE Communications Surveys and telehaptic jitter QoS violation. Tutorials, vol. 5, pp. 28–39, 2003. Based on the analysis in [9], it is reasonable to argue that the [5] S. Floyd, A. Gurtov, and T. Henderson, "The NewReno peak instantaneous haptic jitter increases with the number of modification to TCP’s fast recovery algorithm," 2004. concurrent TCP streams as well as their packet sizes. For QoS [6] R. Wirz, M. Ferre, R. Marin, J. Barrio, J. M. Claver, and J. –compliance, one needs to theoretically determine upper Ortego, "Efficient transport protocol for networked haptics bounds for these two factors. The network administrator then applications," in Haptics: Perception, Devices and needs to ensure that the cross-traffic satisfies the two bounds. Scenarios, Springer, 3–12, 2008. This guarantees satisfaction of QoS conditions for telehaptic jitter. [7] V. Gokhale, J. Nair, and S. Chaudhuri, "Opportunistic adaptive haptic sampling on forward channel in telehaptic communication," in Haptics Symposium, Apr 2016. [8] V. Gokhale, J. Nair, and S. Chaudhuri, "Congestion control for network-aware telehaptic communication," in ACM Transactions on Multimedia Computing, Communications, and Applications, vol. 13, 2017. [9] V. Gokhale, J. Nair, and S. Chaudhuri, "Teleoperation over a shared network: When does it work?," in International Symposium on Haptic, Audio and Visual Environments and Games, Oct 2017. Fig. 6. Variation of peak instantaneous telehaptic jitter with TCP [10] "NS3 - The network simulator", 2011, online - packets of size 1042 B. http://www.nsnam.org/. ACIT 2018, June 1-3, 2018, Ceske Budejovice, Czech Republic