Wireless link layers, which perform retransmissions to hide transmission errors from upper layers, enforce inorder delivery to avoid triggering TCP's congestion control mechanisms. However, new reordering robust TCP flavors make it possible to revisit the design of these link layers. In this paper, we study the effects of the link layer configuration (in-order vs out-of-order) on network layer buffering and transport layer smoothness in a WWAN scenario through simulations. We use a standards-compliant TCP, TCP-Aix, and TCP-NCR. The results show that smoothness is improved and the buffer requirement is reduced when out-of-order delivery is allowed.

Delayed acknowledgments were introduced to conserve network and host resources. Further reduction of the acknowledgment frequency can be motivated in the same way. However, reducing the dependency on frequent acknowledgments in TCP is difficult because acknowledgments support reliable delivery; loss recovery; clock out new segments, and serve as input when determining an appropriate sending rate. Our results show that in scenarios where there are no obvious advantages of reducing the acknowledgment frequency, performance can be maintained although fewer acknowledgments are sent. Hence; there is a potential for reducing the acknowledgment frequency more than is done through delayed acknowledgments today. Advancements in TCP loss recovery is one of the key reasons that the dependence on frequent acknowledgments has decreased.We propose and evaluate an end-to-end solution, where four acknowledgments per send window are sent. The sender compensates for the reduced acknowledgment frequency using a form of Appropriate Byte Counting. The proposal also includes a modification of fast loss recovery to avoid frequent timeouts.

In this paper, we describe and evaluate TCP-Aix, a sender-side TCP algorithm designed to make TCP performance robust to packet reordering and delay. TCP-Aix consists of three novel components: (i) decoupling of loss recovery actions from congestion control actions, (ii) candid growth of the congestion window upon reception of dupacks, allowing for improved performance in reordering environments, and (iii) an algorithm for determining an appropriate dupthresh value. The results from our simulation study show that, by combining these three components, the performance of TCP-Aix is largely impervious even to the highest packet reordering rates investigated in this paper of 8\%, where the duration of the reordering is in the order 2-3 end-to-end RTTs. TCP-Aix outperforms a simulated state-of-the-art TCP sender as well as a TCP-NCR sender, the performance of which both display a much stronger dependence on packet reordering. Improved performance in the presence of delay spikes is also shown. Strengthened by simulation results, we further argue that the improved performance of TCP-Aix does not come at the expense of other, competing, TCP senders. Rather, the TCP-Aix senders utilise bandwidth that would otherwise have been left unutilised. We therefore believe that a gradual deployment of TCP-Aix would be feasible.

In this paper we investigate the influence of buffer management for TCP on performance of the High Speed Downlink Channel (HS-DSCH) introduced in WCDMA release 5. HS-DSCH is a shared channel, but user data is buffered individually prior to the wireless link. Three queue management principles, e.g., passive queuing, the Packet Discard Prevention Counter (PDPC) method and the Random Early Detection (RED) algorithm were evaluated for a number of buffer sizes and scenarios. Also, a buffer large enough to prevent packets from being lost was included for reference. For round robin (RR) scheduling of radio-blocks, PDPC and the passive approach, that both manage to keep the buffer short, gave the best system goodput as well as the shortest average transfer times together with the excessively large buffer. With signal-to-interference ratio (SIR) scheduling, the strategy to avoid all packet losses, resulted in a lower system goodput than for the short buffers. As illustrated in this article, peak transfer rates may not be achieved with very small buffers, but buffers of 10-15 IP packets seem to represent a good trade-off between transfer rates, delay and system goodput. We would like to investigate how to make use of system parameters such as the current amount of data offered for HS-DSCH in total to regulate individual buffer sizes.

We have investigated how the DCCP-thin CCID profile (based on TFRC) for the new transport protocol DCCP behaves in a GPRS network through experimental studies. The results show that performance can be improved by adding two options left out of the DCCP-thin algorithm.

Longest prefix matching is frequently used for IP forwarding in the Internet. Data structures used must be not only efficient, hut also robust against pathological entries caused by an adversary or misconfiguration. In this paper, we attack the longest prefix matching problem by presenting a new algorithm supporting high lookup performance, fast incremental updates and guaranteed compression ratio. High lookup performance is achieved by using only four memory accesses. Guaranteed compression ratio is achieved by combining direct indexing with an implicit tree structure and carefully choosing which construct to use when updating the forwarding table. Fast incremental updates are achieved by a new memory management technique featuring fast variable size allocation and deallocation while maintaining zero fragmentation. An IPv4 forwarding table data structure can be implemented in software or hardware within 2.7 Mb of memory to represent 2/sup 18/ routing entries. Incremental updates require only 752 memory accesses in worst case for the current guaranteed compression ratio. For a hardware implementation, we can use 300 MHz SRAM organized in four memory banks and four pipeline stages to achieve a guaranteed performance of 300 million lookups per second, corresponding to /spl sim/ 100 Gbit/s wire speed forwarding, and 400,000 incremental updates per second. In measurements performed on a 3.0 GHz Pentium 4 machine using a routing table with more than 2/sup 17/ entries, we can forward over 27 million IPv4 packets per second, which is equivalent to wire speeds exceeding 10 Gbit/s. On the same machine and with the same routing table, we can perform over 230,000 incremental updates/second.

This paper explores interactions between congestion control mechanisms at the transport layer and scheduling algorithms at the physical layer in the High-Speed Down-link Packet Access extension to WCDMA. Two different approaches to congestion control - TCP SACK and TFRC - are studied. We find that TCP SACK and TFRC in most respects perform the same way. SIR scheduling give a higher system throughput for both protocols than RR scheduling, but introduces delay variations that lead to spurious timeouts. The no feedback timeout of TFRC was shown to exhibit a similar sensitivity to delay spikes as the retransmit timeout in TCP SACK

With current Internet protocols, users may experience low and unpredictable forwarding quality at wireless links. This is due to varying link properties caused by changing radio conditions. Decreased forwarding quality can cause severe degradation in utilization. This is undesirable since forwarding capacity often is expensive at wireless links because of the limited radio spectrum. Allowing the application and transport layers to communicate with wireless link layers can improve the forwarding quality and utilization. We propose to enable inter-layer communication by adding hints and notifications (HAN) to the Internet architecture. Hints can be introduced and used without notifications, while notifications need hints or a similar mechanism to operate. By using IP options and ICMP messages to implement HAN, a backward-compatible partial deployment is possible. With HAN, the network layer becomes truly wireless friendly and the radio spectrum can be used efficiently while supporting both real-time and traditional data applications.

Over the last decade, cell phones have gone from providing the traditional interactive voice service to also include support for Internet services. Cellular networks that originally were designed for cost-efficient voice services supply low bandwidth, high error rates and long round-trip delays, which poorly fit the generous header sizes, lack of error tolerance and reliance on retransmissions of the Internet. In order to enable cost-efficient Internet connectivity with sufficient service quality in cellular networks, problematic mechanisms in the Internet must be identified and redesigned. The size of the existing Internet calls for solutions that can be deployed only where needed without requiring a major update of the Internet. In this thesis, three specific problems are described together with proposed solutions. Focus is on IP-based voice services, since this is identified as a key application in future cellular networks. The first problem is related to the low bandwidth and high error rates of cellular networks. Real-time applications tend to encode their data with codecs that are in some degree error-tolerant. Interactive cellular voice services use this to a large extent to support usage of data that have been exposed to interference. Using data despite errors is however more problematic in the Internet, since the transport protocol used for this type of service, UDP, discards all datagrams with errors. To address this problem, we introduce a new variant of UDP known as UDP Lite, which provides increased flexibility for applications that prefer partially damaged datagrams over discarded ones. Due to the close relationship between UDP Lite and UDP, UDP Lite is simple to implement. In combination with header compression, UDP Lite enables cost-efficient IP-based voice services in a cellular phone network. The second problem is related to the introduction of new header compression algorithms, which suffers from long delay due to standardization and deployment. This prevents efficient compression of application-level headers that can be large. Even if headers can be compressed in theory, the delay until the algorithm is widespread is too high. We propose a framework in which header compression algorithms can be expressed with platform-independent descriptions. Descriptions of header compression algorithms are dynamically installed and removed when needed. These descriptions are retrieved either from the application, the link peer, or from a well-known server. The third problem addresses a specific problem in the design of the Internet protocol stack - the lack of communication between the link layer and upper layers. With knowledge about the properties of the data in a link layer frame, expensive wireless bandwidth in cellular systems could be used more efficiently. This problem has been addressed in many papers that focus on particular applications or protocols. We propose a generic system called HAN that enables cross-layer interaction to support intelligent decision-making where needed. By sending hints to, and getting notifications from, the link layer, better performance can be obtained in parts of the Internet where HAN is supported. The solutions proposed in this thesis all aim at being generic and backward compatible with existing Internet technology. The UDP Lite protocol is currently being standardized within the IETF. The combination of UDP Lite and header compression is likely to exist in upcoming 3G networks. The work with a header compression framework and cross-layer interaction is still in progress.