We first explain problems with the default scheduler used by the Linux kernel MPTCP implementation. Then we propose the design of a new scheduler. Preliminary empirical results show that our proposed scheduler improves the throughput in MPTCP by alleviating the problems caused by the default scheduler.

In both TCP and SCTP, selectively acked (SACKed) out-of-order data is implicitly renegable; that is, the receiver can later discard SACKed data. The possibility of reneging forces the transport sender to maintain copies of SACKed data in the send buffer until they are cumulatively acked. In this paper, we investigate the situation where all out-of-order data is non-renegable, such as when the data has been delivered to the application, or when the receiver simply never reneges. Using simulations, we show that SACKs result in inevitable send buffer wastage, which increases as frequency of loss events and loss recovery durations increase. We introduce a fundamentally new ack mechanism, Non-Renegable Selective Acknowledgments (NR-SACKs), for SCTP. Using NR-SACKs, an SCTP receiver can explicitly identify some or all out-of-order data as being non-renegable, allowing the sender to free up send buffer sooner than if the data were only SACKed. Simulation comparisons show that NR-SACKs enable efficient utilization of a transport senderpsilas memory. Further investigations show that NR-SACKs also improve throughput in Concurrent Multipath Transfer (CMT) [4].

We propose using the Stream Control Transmission Protocol (SCTP), a recent IETF transport layer protocol, for reliable web transport. Although TCP has traditionally been used, we argue that SCTP better matches the needs of HTTP-based network applications. This position paper discusses SCTP features that address: (i) head-of-line blocking within a single TCP connection, (ii) vulnerability to network failures, and (iii) vulnerability to denial-of-service SYN attacks. We discuss our experience in modifying the Apache server and the Firefox browser to benefit from SCTP, and demonstrate our HTTP over SCTP design via simple experiments. We also discuss the benefits of using SCTP in other web domains through two example scenarios ? multiplexing user requests, and multiplexing resource access. Finally, we highlight several SCTP features that will be valuable to the design and implementation of current HTTP-based client-server applications.

Previously, we studied the performance of concurrent multipath transfer using SCTP multihoming (CMT) under the assumption of an infinite receive buffer (rbuf). Here, we study CMT performance when a sender is constrained by the rbuf. We demonstrate using simulation that if two paths are used for CMT, the lower quality (i.e., higher loss rate) path degrades overall throughput of an rbuf-constrained CMT association by blocking the rbuf. We demonstrate that a wise retransmission policy can alleviate some of the throughput degradation by reducing the rbuf blocking problem. We present and discuss CMT performance using several retransmission policies and constrained rbuf values of 16 KB, 32 KB, 64 KB, 128 KB, and 256 KB. While rbuf blocking cannot be eliminated, it can be reduced by choice of retransmission policy - a facility available to only the transport layer.

We previously evaluated five retransmission schemes in non-failure scenarios for transport protocols that support multihoming. We now introduce five additional retransmission schemes, and evaluate all ten schemes under both non-failure and failure scenarios. We show that the best retransmission policy dictates that (a) new data transmissions and fast retransmissions should be sent to the same peer IP address, and (b) timeout retransmissions should he sent to an alternate peer IP address. This policy performs best if combined with our multiple fast retransmit algorithm.

In recent years several enhancements to TCP congestion control and loss recovery mechanisms have been proposed and accepted as Internet standards. In this paper, we examine these enhancements in detail. We consider five TCP enhancements: (1) selective acknowledgements (SACK) and the SACK-based loss recovery algorithm; (2) increasing initial congestion window; (3) limited transmit; (4) appropriate byte counting; and (5) early retransmit. We present results from active measurements performed on web servers on the state of deployment of these enhancements. Our results show that while several web servers support TCP enhancements, the majority still use previous standards for congestion control and loss recovery. Using simulation, we quantify the cumulative effect of these TCP enhancements on web based and bulk data transfers. We hope that such an evaluation provides a clearer view of the applicability of these enhancements, and further motivation for their implementation.

We identify overheads associated with FTP, attributed to separate TCP connections for data and control, non-persistence of the data connections, and the sequential nature of command exchanges. We argue that solutions that avoid these overheads using TCP place an undue burden on the application. Instead we propose modifying FTP to use SCTP and its multistreaming service. FTP over SCTP avoids the identified overheads in the current FTP protocol without introducing complexity at the application, while still remaining "TCP-friendly". We implemented FTP over SCTP in three ways: (1) simply replacing TCP calls with SCTP calls, thus using one SCTP association for control and one SCTP association for each data transfer, (2) using a single multistreamed SCTP association for control and all data transfers, and (3) enhancing (2) with the addition of command pipelining. Our experiments compared these 3 variations with the classic FTP over TCP. Results indicate significant improvements in throughput for multiple file transfers with all three of our variations. The largest benefit occurs for (3) FTP over a single, pipelined, multistreamed SCTP association. More generally, this paper encourages the use of SCTP's innovative services to benefit existing and future application performance and presents the case for multistreaming.

We present a new model for testing real-time protocols with multiple timers, which captures complex timing dependencies by using simple linear expressions involving timer-related variables. This new modeling technique, combined with the algorithms to eliminate inconsistencies, allows generation of feasible test sequences without compromising their fault coverage. The model is specifically designed for testing to avoid performing full reachability analysis, and to control the growth of the number of test scenarios. Based on extended finite state machines, it is applicable to languages such as SDL, VHDL, and Estelle. The technique models a realistic testing framework in which each I/O exchange takes a certain time to realize and timers can be arbitrarily started or stopped. A software tool implementing this technique is used to generate test cases for the US Army wireless standard MIL-STD 188-220.

This paper presents results from performance experiments that demonstrate and quantify performance improvements when a PO/R transpor5t service is used instead of an ordered/reliable service (O/R e.g., TCP) or an unordered/unreliable service (e.g. UDP). We first describe the Remote Multimedia Document Retrieval system (ReMDoR), an experimental application developed by the authors to evaluate the performance of remote document retrieval over a variety of transport protocols. We then provide a detailed analysis of experiments comparing O/R service to PO/R service for retrieval of a multimedia document. Our results show that between 5% and 10% loss, user-perceivable improvements in progressive display are obtained when PO/R service is used. These results suggest that when packet losses occur in an underlying packet-switched network, transport services providing reliable delivery over independent streams (such the emerging Internet protocol SCTP) are beneficial for retrieval of streaming multimedia.

Wavelet zerotree encoding has been proven to be an efficient way of compressing still images. Two well-known zerotree encoding algorithms, embedded zerotree encoding (EZW) and set partitioning in hierarchical trees (SPIHT), provide excellent progressive display when images are transmitted over reliable networks. However, both algorithms are state-dependent and can perform poorly over unreliable networks. In this paper, we apply the concept of network-conscious image compression to the SPIHT wavelet zerotree encoding algorithm, to improve its performance over unreliable networks. Experimental results confirm the utility of network-conscious image compression concept.

An analytic model is presented for a partially reliable transport protocol based on retransmissions. The model illustrates tradeoffs between two QoS parameters (delay and throughput), and various levels of reliability. The model predicts that the use of a reliable transport service when an application only needs a partially reliable one causes considerable throughput decreases and delay increases in lossy networks. On the other hand, over lossy networks, an unreliable transport service is unable to respect an application's loss tolerance. In lossy environments, partially reliable transport service avoids the extra cost of a reliable transport service, and, simultaneously, guarantees the minimal reliability that an application requires. A retransmission-based partially reliable transport service can be provided through either sender-based or receiver-based loss detection and recovery. Results show that both techniques provide almost identical reliability and delay. However, a sender-based approach provides better throughput than a receiver-based approach at high acknowledge loss rates.

Many applications such as video and audio can tolerate loss. When the network layer provides a best-effort service such as on the Internet, the loss rate of the underlying network service may be higher than an application's tolerance for loss. This paper analytically studies retransmission-based partially reliable transport (layer protocol) service. Results show that a partially reliable transport service provides increasingly higher throughput and lower delay than a reliable transport service as an application's loss tolerance increases and as the underlying network service gets more lossy. Also, to some degree, a partially reliable transport service eases the negative effects of ack losses on throughput. Three cost functions associated with the reliability level that a system can support are introduced. These cost functions help demonstrate the penalty when a transport service does not support the ideal reliability level for an application. Results show that the use of a reliable transport service when an application only needs a partially reliable transport service can cause considerable throughput drops and delay increases in lossy networks. On the other hand, at high loss rates, an unreliable transport service is unable to respect an application's loss tolerance. Thus, in lossy environments, a partially reliable transport service is necessary to avoid the extra cost of a reliable transport service, and, at the same time, to guarantee the minimal reliability that an application requires.

Investigates a partial-order connection (POC) service/protocol. Unlike classic transport services that deliver objects either in the exact order transmitted or according to no particular order, POC provides a partial-order service, i.e. a service that requires some, but not all objects to be received in the order transmitted. Two versions of POC are proposed: reliable, which requires that all transmitted objects are eventually delivered, and unreliable, which permits the service to lose a subset of the objects. In the unreliable version, objects are more finely categorized into one of three reliability classes depending on their temporal value. Two metrics based on e/sub i/(P), the number of linear extensions of partial-order P in the presence of i lost objects, are proposed as complexity measures of different combinations of partial order and reliability. Formulae for calculating e/sub i/(P) are derived when P is series-parallel. A formal specification of a POC protocol, written in Estelle, is presented and discussed. This specification was designed and validated using formal description tools and provides a basis for future implementations.>

This paper introduces a partial order connection (POC) protocol. Motivated in particular by multimedia applications, POC is an end-to-end connection that provides a partial order service, that is, a service that requires some, but not all objects to be received in the order transmitted. This paper discusses R-PO, a reliable version of POC which requires that all transmitted objects are eventually delivered. A metric based on the number of linear extensions of a partial order in the presence of no lost objects is proposed to quantify different partial orders. Means for its calculation is presented when P can be modeled as a combination of sequential and/or parallel compositions of Petri-nets. This metric allows one to compare and evaluate the complexity of different partial order services.>

Using notation consistent with the Estelle ISO 9074 International Standard, the syntax and formal semantics for a broadcast channel are presented. These semantics make precise the meaning of an intuitive concept and additionally are compatible with those of the current Estelle FDT (formal description technique). If introduced to the Estelle FDT, broadcast channels would strengthen its specification capabilities by permitting clearer formal specifications. Broadcast channels provide a foundation for a powerful synchronization mechanism.>

Using simulation, it is shown that two changes to the fiber distributed data interface (FDDI) protocol suggested by K.C. Sevcik and M.J. Johnson (1987) will simplify the protocol and improve the performance of synchronous traffic by reducing its mean latency. The first change requires that an asynchronous frame must be able to be sent in its entirety in the time remaining before the token rotation timer expires. The second change causes the token rotation timer to be reset to zero on every arrival, including late ones, thereby discarding token lateness instead of accumulating it.>

An overview is presented of the statistical sampling of measurements from a computer network, a necessary task in all five facilities of the proposed ISO network management architecture. Network measurement, viewed from the standpoint of real-time status reporting, is performed either to approximate the condition of a network parameter over a given interval or to detect a significant change in system behavior over consecutive intervals. Appropriate statistical tools for sampling and analyzing traffic in real time are presented.>

An architecture model and a subset of a behavior model of the ISO virtual terminal (VT) are specified in the formal description technique Estelle. The architecture model includes modules and their appropriate interconnections (channels) for a VT-user, a VT protocol machine (VTPM), an association control service entity (ACSE), autonomous-abort internal events, a presentation layer element, and a mapping machine to map protocol data units of the VTPM onto service primitives of the ACSE and the presentation layer. The behavior model specifies the association establishment phase of the VTPM for the synchronous mode of operation, assuming that no optimal function units are selected. The combination of VT and Estelle represents ISO's effort toward implementing the Open Systems Interconnection (OSI) model.>

(This paper has been accepted for publication in the Proceedings, but the photo-ready form was not delivered in time. Copies of the paper should be available upon request at the presentation.)

This paper describes a measurement center for the NBSNET, a distributed, broadcast local area computer network (LAN) at the National Bureau of Standards. A LAN measurement center allows careful testing and evaluation of a network under normal and varying user-defined conditions. The measurement center consists of three components: an artificial traffic generator, a monitoring system, and data analysis software. The traffic generator emulates varied loads on the network, allowing for controlled experimentation and functional testing. The monitoring system captures measurement information about both artificial and normal network traffic. Analysis software summarizes this information into ten measurement reports following each monitoring period. Implementation issues and problems are discussed.