(Technical report) Jason Nieh, S. Jae Yang, Naomi Novik, “A Comparison of Thin-Client Computing Architectures”, Technical Report CUCS-022-00 November 2000. Techniques were developed to analyze thin clients performance to assess the general feasibility of the thin-client computing model and comparing different thin-client platforms and determine the factors that govern performance. The platforms that were assessed were Citrix Mataframe, Microsoft Windows Terminal Server, Lablink 2000, AT&T VNC and Sun Microsystem Sun Ray. Emphasis was placed on popular Web multimedia applications. The results indicated that current thin-client platforms are viable alternatives for a typical LAN environment or 10Mbps. However, the thin-clients were lacking with multimedia-intensive workloads over wide area networks. The results also showed that higher-level encodings are not necessarily ideal for graphics-intensive multimedia applications with many rapidly changing images.
(Conference Article) W. F. Mann, et al. “A network-oriented multiprocessor front-end handling many hosts and hundreds of terminals”, Proceeding AFIPS '76 Proceedings of the June 7-10, 1976, national computer conference and exposition. DOI: http://dx.doi.org/10.1145/10.1145/1499799.1499875 A design of a network system with multiple terminals and hosts. The system allows for multiple terminals to be attached to multiple hosts, for the hosts to communicate with one another and for the host to share access to I/O devices. Design issues considered included; dedicated function machines, terminal handling and service functions were dealt with on the same machine that store/forward IMP functions, terminal interface hardware. The ability to connect at least three different types of terminals to an system port. The communication protocol, which protocols should terminals use to communicate to the host and host to host communication. The user’s interface to the host, this included the removal of double login to TIP and then the host, automatic connection to an available host and the ability to switch host while login into another host. The remaining design issue were interfacing with remote host and, flow control and buffering among terminals and host.
(Conference Article)S. M. Ornstien et.al. “The terminal IMP for the ARPA computer network”, Proceeding AFIPS '72 (Spring) Proceedings of the May 16-18, 1972, spring joint computer conference. DOI: http://dx.doi.org/10.1145/10.1145/1478873.1478906 Description of a communications network extending the ARPA network to permit resource sharing, reliable and economic digital communications and to permit broad access to unique facilities. Messages transverse several nodes, each message carries the destination address and a copy of the message is stored at each node until it is safely received at the following node. Each node has an Interface Message Processor (IMP) which acts as a switching unit. Distribution of computational power among the remote big facilities, terminal processors, and terminals were considered. The decision was made to keep terminal processors simple and not programmable with computational load and storage in the host. Another major design issue was to separate the terminal handler from the IMP. In anticipation of higher bandwidths and saturation of future workloads the design of a significantly faster and modular version of IMP was considered.
(Journal Article) Scheifler, Robert W., et al. “The X Window System,” ACM Transactions on Graphics, vol. 5, no. 2, pp. 79-109, April, 1986. DOI: http://dx.doi.org/10.1145/10.1145/10.1145/22949.24053 X windows is a graphical system allowing applications to utilize windows on any display in a network in a device independent, network-transparent fashion. A based system is defined by a network protocol: asynchronous stream-based inter-process communication. Both graphics functions and “system” functions are pushed back (across the network) as application functions, and in the ability to tailor desktop management transparently. The X-window system is based on client-server model. Performance is optimized principally by minimizing the number of operating system calls. The server and network protocols are robust enough to never trust clients to provide correct data. Inter-process round trip communication time is between 5 and 50 milliseconds.
(Dissertation) Nowicki, W. “Partitioning of function in a distributed graphic system” Ph.D. dissertation, Department Computer Science, Stanford University, CA 1985. This thesis discusses the trade-offs involved in partitioning of function in a distributed graphics system. Performance is one important property traded for advanced functionally or decreased cost. To provide adequate performance in a distributed system, communication costs should be kept low, as well as the frequency of the communication. By providing modeling as well as viewing facilities, the VGTS(Virtual Graphics Terminal Service) reduces the communication required between applications and the service.
(Conference Article) Schmidt, Brian K., et al “.The interactive performance of SLIM: a stateless, thin-client architecture” Proceeding SOSP '99 Proceedings of the seventeenth ACM symposium on Operating systems principles. DOI: http://dx.doi.org/10.1145/10.1145/10.1145/319344.319154. SLIM: a stateless, low-level interface machine providing resource sharing, centralized administration and inexpensive units. It is a simple pixel encoding protocol requiring only modest network protocol. The workgroup environment only focuses on SLIM consoles connected to servers via a private network carrying only SLIM protocol traffic. The SLIM architecture is composed of an interconnection fabric, a private communication medium with dedicated connections between the desktop units and the serves. The SLIM protocol represents the lowest common denominator of all rendering API’s by encoding raw pixel values. The protocol also consists of a small number of messages for communicating status between desktop and server. Evacuations showed that high-demand multimedia and live data applications are supported by the SLIM architecture and the server performance was the primary bottleneck. With 3D rendering the performance limiting factor was the client CPU.
Conference Article) Conner, B, et al. “Providing a low latency user experience in a high latency application”, Proceeding I3D '97 Proceedings of the 1997 symposium on Interactive 3D graphics. DOI: http://dx.doi.org/10.1145/253284.253305 A low latency user experience, even when extremely large latencies occur in an application. During latency lags visual effects such as motion blur, transparency, and defocusing are used to provide immediate feedback. When latency lag is removed the same effects are used to provide coherent updates. The effects are constructed as new Open Inventor node types using Open GL directly to render the effects. This lets these effects be used in a wide variety of applications, while providing extremely good performance.
(Conference Article) Nieh, J., et al. “The design, implementation and evaluation of SMART: a scheduler for multimedia applications”, Proceeding SOSP '97 Proceedings of the sixteenth ACM symposium on Operating systems principles. DOI: http://dx.doi.org/10.1145/269005.266677. SMART, a Scheduler for Multimedia and Real-time applications and provides dynamic feedback to applications to allow them to adapt to the current load. SMART was implemented in the Solaris UNIX operating system and measured its performance against 0ther schedulers in executing real-time, interactive, and batch applications The results showed reduced burden of writing adaptive real-time applications, cooperated with applications in managing resources to meet their dynamic time constraints, provided resource sharing across both real-time and conventional applications, delivered improved real-time and interactive performance over other schedulers without any need for users to reserve resources, adjust scheduling parameters, or know anything about application requirements, provided flexible, predictable controls to allow users to bias the allocation of resources according to their preferences
(Technical report) Boca Research, Inc., “Citrix ICA Technology Brief”, Technical white Paper, Boca Raton, FL, 1999. ICA is thin-client software in conjunct ion with an application server and a distributed Windows display protocol. It is a network-centric, three-tiered application architecture where the application execution and data storage occurs on a central server (or servers), and only a thin piece of client software is required at the client system. ICA uses industry standard network protocols to provided high-performance Windows display over low bandwidth connections.
(Conference Article) Albert Lai, et al. ”Limits of wide-area thin-client computing”, Proceeding SIGMETRICS '02 Proceedings of the 2002 ACM SIGMETRICS international conference on Measurement and modeling of computer system pages 228–239, 2000. DOI: http://dx.doi.org/10.1145/511399.511363 Evaluation of thin client computing in a wide area network (WAN) by using a non-invasive slow motion bench mark technique. Evaluation has shown that performance varies among platforms and that network latency is often a limiting factor. Sun Ray delivered excellent performance on all of the benchmarks measured. Implications of the evaluations produced five principles that should serve as a guideline in designing thin-client for WAN: optimizing latency over bandwidth, partition client/server functionality to minimize synchronization, use simpler display primitives for speed, compress display updates, and push display updates.
(Conference Article) Wong, A.Y., et al. “Operating system support for multi-user, remote, graphical interaction”, Proceeding ATEC '00 Proceedings of the annual conference on USENIX Annual Technical Conference. Investigates the impact of operating system design on the performance of thin client service. The key performance metric used was user-perceived latency and an analysis of the performance and scalability of thin client server operating systems, Windows NT, Terminal Server Edition and Linus X Windows were made. The investigation revealed that resource scheduling for both the processor and memory in these systems is not well optimized for heavy concurrent interactive use. Latency and jitter was well above human-perceptible levels. The RDP protocols was generally more efficient that X and LBX in terms of network loads.
(Conference Article) Fok, Federic A.C., et al. “An adaptive approach to optimize thin client protocols”, Future Network & MobileSummit 2010 Conference Proceedings Paul Cunningham and Miriam Cunningham (Eds) IIMC International Information Management Corporation, 2010. Adaptive protocols were presented in both upstream and downstream direction to deal the varying characteristics of wireless networks and to reduce energy consumption of thin client mobile devices. Evaluation of the adaptive protocols showed that n the downstream direction, an adaptive and scheduled-based screen update transmission pattern was introduced. Benefits included reduced bandwidth and energy consumption with still a good or acceptable QoE. In the upstream direction, the buffering of user events decreases the packetization overhead and leads to important bandwidth savings. The buffering period is adapted to the current network delay, in order to preserve the user perceived latency.
(Conference Article) Chung, Y. et al. “Understanding the performance of thin-client gaming”, Communications Quality and Reliability (CQR), 2011 IEEE International Workshop Technical Committee on. DOI: http://dx.doi.org/10.1109/CQR.2011.5996092. Methodology for quantifying the performance of thin-clients on gaming. Three thin clients were evaluated; LogMeIn, TeamViewer, and UltraVNC. Results showed that the display frame rate and frame distortion of the client are both critical to gaming performance. Also, different thin-client implementations may have very different levels of robustness against network impairments. While different implementations may excel in different network situations, LogMeIn in general performs the best among the three systems studied.
(Conference Article) Simoens, P. et al. “Characterization of power consumption in thin clients due to protocol data transmission over IEEE 802.11”, Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks, 2009. WiOPT 2009. 7th International Symposium. DOI: http://dx.doi.org/10.1109/WIOPT.2009.5291601. Experimental results on power efficiency of the wireless platform on the thin client to demonstrate whether or not the energy needed for the connection between the client and server might undo or even exceed the power savings achieved by the reduction in client-side processing. Results showed that depending on the channel conditions and requirements, a reduction of the power consumption of 14% can be achieved. More energy efficiency can be obtained, by reducing time spent in the idle state, by introducing dynamic sleep time schedules for the wireless platform. In this respect, the Automatic Power Save Delivery (APSD) in the 802.11e specification seemed highly relevant.
(Conference Article) Tan, K., et al. “A remote thin client system for real time multimedia streaming over VNC”, Conference ProceedingsMultimedia and Expo (ICME), 2010 IEEE International July 2010. DOI: http://dx.doi.org/10.1109/ICME.2010.5582993. A remote thin client system for VNC that increases frame rates and video quality under bandwidth limitation. The proposed thin client system has a structure which splits frames into low motion and high motion parts. To ensure that the video can be transmitted in real time, RTP/UDP instead of TCP is used to transmit high motion part of frame. Since UDD is unreliable a Error Detection Scheme is used to make sure the decoding process will not be terminated by reason of packet lost. Results showed the proposed thin client system; over 22 fps of display rate is achieved under the limitation of 32KByte/s bandwidth, which is about 172 times faster as compared to pure VNC for remote high motion video. The performance of the proposed system is limited by the adopted H.264 video encoding card, which could be further enhanced if more powerful H.264 video encoder is adopted.
(Conference Article) Grundy, J., et al. “An environment for automated performance evaluation of J2EE and ASP.NET thin-client architectures”, Software Engineering Conference, 2004. Proceedings. 2004 Australian. DOI: http://dx.doi.org/ 10.1109/ASWEC.2004.1290483. Description of a thin client test-bed generator called SofArch/Thin that synthesizes performance test bed thin-client and server code from high-level software architecture models. SofArch/Thin is an extension of earlier work which included developing code generators for both J2EE and C#, ASP.NET thin client application implementation technologies. SofArch/Thin is used to assess the likely performance of thin-client architecture designs. SoftArch/Thin was used to model several thin-client architectures and obtain realistic performance measures via generated test bed code
(Conference Article) Winter, D. , et al. “A hybrid thin-client protocol for multimedia streaming and interactive gaming applications”, Proceeding NOSSDAV '06 Proceedings of the 2006 international workshop on Network and operating systems support for digital audio and video. DOI: http://dx.doi.org/1 10.1145/1378191.1378210. A thin-client system with a real-time desktop-streamer using a videocodec to stream the graphical output of applications after GPU processing. The purpose of the desktop-streamer is to supplement the lack of potential use of the GPU by the thin client. Results from comparison of this approach to classic thin-client systems which are best suited, especially if processing overhead is taken into account, for office programs and web browsing. When multimedia or 3D games are used, the real-time desktop-streamer outperformed classic thin-client systems and is the only viable solution.
(Conference Article) Deboosere, L., et al. “Thin Client Computing Solutions in Low- and High-Motion Scenarios”, Proceeding ICNS '07 Proceedings of the Third International Conference on Networking and Services. DOI: http://dx.doi.org/10.1109/ICNS.2007.115. Tests to show the performance of widely used thin client protocols, RDP, ICA and Sun Ray, in low-motion and high motion scenarios. Results showed that in low-motion all tested thin clients delivered the necessary performance. High-motion scenarios increased network bandwidth and not all protocols produced high quality output. While Citrix outperformed the rest a codec needs to be installed at the client for each type of stream which can result in large number of codecs a luxury thin clients can’t accommodate.
(Conference Article) Simoens, P, et al.. “Design and implementation of a hybrid remote display protocol to optimize multimedia experience on thin client devices”, Telecommunication Networks and Applications Conference, 2008. ATNAC 2008. Australasian. DOI: http://dx.doi.org/10.1109/ATNAC.2008.4783356. A hybrid remote display protocol approach leveraging existing Remote FrameBuffer protocol of Virtual Network Computing (VNC-RFB) protocol with a video streaming mode to transport the rendered images of multimedia applications to the client. Dependent on the amount of motion in the images to be presented, the images are relayed to the client either through the VNC-RFB protocol or through video streaming in the H.264 format. Results showed that video encoding incurs high server and client CPU load. Therefore, the video mode should only be activated when it is really necessary
(Conference Article) Vankeirbilick, B., et al. “Bandwidth Optimization for Mobile Thin Client Computing through Graphical Update Caching”, Telecommunication Networks and Applications Conference, 2008. ATNAC 2008. Australasian. DOI: http://dx.doi.org/10.1109/ATNAC.2008.4783355. Caching as a mechanism to reduce the network load generated by thin client computing systems. The cache proposed in this article is static, meaning that it is composed before the thin client computing session starts and that the cache does not change during the session. The static cache can reduce the generated network traffic by about 34.40%. It is shown that on a momentary basis the generated network traffic decreases on average by 35.17%, with a maximum of 99.81%. Furthermore, ideas are presented on how to select cache frames and how the number of cache frames can be deduced.
(Conference Article) Schlosser, D, et al.. “Performance comparison of windows-based thin-client architectures”, Telecommunication Networks and Applications Conference, 2007. ATNAC 2007. Australasian. DOI: http://dx.doi.org/10.1109/ATNAC.2007.4665260. The Windows remote desktop protocol (RDP) and the Citrix presentation server were compared. The Quality-of-Experience (QoE) perceived by the end-user, comparison of the overhead required by the different available protocols, and their advantages and disadvantages were evaluated. Results showed that Since Citrix is optimized for Microsoft Office applications; it achieves a slightly better Quality-of-Experience under perfect network conditions.
(Conference Article) Emmert, B., et al. “Source traffic characterization for thin client based office applications”, Proceeding EUNICE'07 Proceedings of the 13th open European summer school and IFIP TC6.6 conference on Dependable and adaptable networks and services. Characterization and analysis of the traffic generated by different types of thin client users when working with popular office applications like Microsoft Word, Excel, or PowerPoint. Results showed that the user input, like keystrokes or mouse movements, is mainly encoded with a payload of 6 byte. Thereby higher bandwidth rates are mainly achieved by decreasing the time between two packets while keeping the size of the packets constant. This indicated that in the design of the underlying ICA protocol, responsiveness played a more important role than efficient bandwidth usage. Also, keyboard input causes less traffic than mouse movements or the transfer of large files, but that PowerPoint animations consume comparatively small amounts of bandwidth.
(Journal Article) Tolia, N., et al. “Quantifying Interactive User Experience on Thin Clients” Journal Computer, Volume 39 Issue 3, pages46-52, March 2006. DOI: http://dx.doi.org/10.1109/MC.2006.101. Description of stateless thick clients that preserves many of the benefits of thin-client computing but eliminates its acute sensitivity to network latency. The thick client uses virtual machine technology to enables clean encapsulation of a user’s entire personal computing state. A distributed file system can store VM state and deliver it on demand to any thick client. This approach trades off start-up delay for crisp interaction: Once execution begins, all interaction is local. A thick-client approach assumes that clients have sufficient resources (disk, CPU, and so forth) to run the desired applications locally. The stateless thick-client approach further depends on high network bandwidth for VM state transfer
(Journal Article) Neigh J., et al. “Measuring thin-client performance using slow-motion benchmarking”, Journal ACM Transactions on Computer Systems Volume 21 Issue 1, pp. 87-115, February 2003. DOI: http://dx.doi.org/10.1145/592637.592640. A Benchmarking technique to measure user-perceived performance on the client as opposed to the application performance on the server.
(Conference Article) Gutwin, C., “The effects of network delays on group work in real-time groupware”, Proceeding ECSCW'01 Proceedings of the seventh conference on European Conference on Computer Supported Cooperative Work. The effects of network delays, latency and jitter) closely-coupled with group work in real-time distributed groupware. Relationships between delay and task performance were found.
