Enhancing Fast TCP’s Performance Using Single TCP Connection for Parallel Traffic Flows to Prevent Head-of-Line Blocking

Sarfraz Ahmad,Muhammad Junaid Arshad

doi:10.1109/access.2019.2946527

Abstract

Hypertext Transfer Protocol-2 (HTTP/2) partially resolved the problem of Head-of-Line Blocking (HoLB) by multiplexing independent messages at the application layer. This enables simultaneous transmission of multiple requests over the same connection independently. However, this technique becomes ineffective when packet loss occurs in the Transmission Control Protocol (TCP) flow in which case all the independent streams are blocked until the retransmission of the lost packet; this problem is known as HoLB at TCP-level. The problem arises because the underlying TCP does not differentiate between independent messages/streams from application layer protocol. This study proposes a multistream framework for Fast TCP to support multiple independent messages/streams of the application. The proposed framework uses separate flows, buffers, and segments for each independent stream, and interleaves these segments over a single TCP connection. It makes TCP compatible with HTTP/2, reduces data delivery latency between transport and application layer, and alleviates head-of-line blocking. We implemented this framework in Fast TCP and carried out a simulation-based comparison between Multistream Fast TCP (MFast TCP) and Stream Control Transmission Protocol (SCTP). Our results show that MFast TCP significantly improved performance over SCTP in the event of HoLB.

Highlights

Latest broadband facility is available to home users via modem and ADSL
Our results show that when Head-of-Line Blocking (HoLB) occurs, the MFast Transmission Control Protocol (TCP) performs better compared to the Stream Control Transmission Protocol (SCTP)
We study the performance of MFast TCP in comparison with Fast TCP and SCTP

Summary

INTRODUCTION

Latest broadband facility is available to home users via modem and ADSL. It enables the recent applications to transfer a variety of data to the end user. It provides multistream feature to address the head-of-line blocking problem at the transport layer It efficiently handles small flows of signaling messages, its loss-based congestion control algorithm degrades its performance in high bandwidth-delay product networks [22]. HTTP/1.1 resolved the problems faced by HTTP/1.0 and proposed to open limited persistent TCP connections It reduces overhead like number of network packets, resource requirement on the server, and TCP connection establishment over multiple requests. It multiplexes data of all the requests over a single TCP connection This technique eliminates the need for multiple TCP connections, it could not remove the head-of-line blocking problem at the transport layer [11], [12]. Its multistream feature addresses the problem of head-of-line blocking, its loss-based congestion control algorithm degrades its performance over long distance, large bandwidth networks. The Data Acknowledgment (DA) unit acknowledges each received segment on the receiver side and coordinates with the Data Manager (DM) on the sender side to identify any lost segments so that the Data Manager could adjust its data selection

IMPLEMENTATION DETAIL

STREAM HEADER

35 Dispatch ACK Segment

ENVIRONMENT AND ASSUMPTION

Findings

CONCLUSION