Pasi Sarolahti

Spurious TCP retransmission timeouts are a considerable problem degrading the TCP efficiency in communication over mobile 2.5G and 3G networks. They cause several unnecessary retransmissions and diminish TCP throughput. F-RTO algorithm can be used to detect spurious TCP RTOs, thus making it possible to avoid unnecessary retransmissions in such occasions and improve TCP performance. In this paper we present a SACK-based enhancement to the F-RTO algorithm that further improves the efficiency of F-RTO in detecting spurious RTOs. Additionally, we evaluate three congestion control alternatives to be applied when spurious TCP RTO is detected under different levels of congestion. The results show that choosing the most efficient congestion control alternative depends on the available link bandwidth, amount of buffering space in the bottleneck router, and the number of parallel TCP connections. If the wireless link has a low bandwidth, sufficient conservativeness at the sender not only prevents congestion losses, but also improves TCP throughput. On a faster link more aggressive congestion control yields better throughput despite of increased number of congestion losses.

... The results show that our algorithm either improves performance or gives similar throughput as the other TCP variants evaluated in different test cases. 1. INTRODUCTION In the recent years the variety of Internet links with differ-ent properties has increased dramatically. ...

... The results show that our algorithm either improves performance or gives similar throughput as the other TCP variants evaluated in different test cases. 1. INTRODUCTION In the recent years the variety of Internet links with differ-ent properties has increased dramatically. ...

Vertical hand-offs between different wireless access technologies have become more relevant after the introduction of multi-access mobile terminals with wireless LAN (WLAN) and wireless WAN (WWAN) technologies. While the IP mobility mechanisms are rather well known, the performance of TCP still has problems when moving between WLAN and WWAN accesses. First, with a high-latency WWAN link technology such as GPRS it takes several seconds before the TCP congestion window has reached the path capacity. Second, when the notification of the first packet loss arrives at the TCP sender, several packets have already been lost due to the slow-start overshoot and the TCP sender needs to retransmit a large number of the packets from the last transmission window. Third, after a vertical hand-off the path characteristics might have changed dramatically in which case the TCP congestion control state is not valid any more. In this paper we investigate Quick-Start, a mechanism for avoiding the initial slow-start delay, in the context of wireless multi-access terminals. We also propose an enhancement to Quick-Start to alleviate the effects of slow-start overshoot and apply Quick-Start after a vertical hand-off to quickly learn the available capacity on the new end-to-end path. An explicit cross-layer hand-off notification is employed to trigger Quick-Start when the hand-off completes. We conduct simulations with different hand-off models, and our simulations yield promising results with Quick-Start

[Docs] [txt|pdf] [draft-ietf-tsvwg-...] Diff1] [Diff2] [Errata] EXPERIMENTAL Errata Exist Network Working Group S. Floyd Request for Comments: 4782 M. Allman Category: Experimental ICIR A. Jain F5 Networks P. Sarolahti Nokia Research Center January 2007 Quick-Start for TCP ...