Scribd
Upload
81 views

Practical Experience With HTTP and TCP Over GPRS

This document discusses improving web performance over GPRS networks. It summarizes that: 1. GPRS links have high round-trip times (RTTs) of over 1000ms and occasional outages of 4-40 seconds can occur. 2. A mobile proxy located near the wireless-wireline border and enhancements to TCP and applications can improve performance over GPRS. 3. Experimental tests showed that using a performance-enhanced proxy with moderate HTTP pipelining in a web browser reduced mean page download times by 15-20% over GPRS.

Uploaded by

Eithu Thutun
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
81 views

Practical Experience With HTTP and TCP Over GPRS

This document discusses improving web performance over GPRS networks. It summarizes that: 1. GPRS links have high round-trip times (RTTs) of over 1000ms and occasional outages of 4-40 seconds can occur. 2. A mobile proxy located near the wireless-wireline border and enhancements to TCP and applications can improve performance over GPRS. 3. Experimental tests showed that using a performance-enhanced proxy with moderate HTTP pipelining in a web browser reduced mean page download times by 15-20% over GPRS.

Uploaded by

Eithu Thutun
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

Practical Experience with HTTP and TCP over GPRS

Rajiv Chakravorty and Ian Pratt, Cambridge University, U.K. GPRS Link Characterization Improving Performance: How?

+ Use an interposed Mobile Proxy, located close to + High RTTs: 1000ms or more wireline-wireless border near GPRS CGSN Node + Links Outages: Typically, observed for duration of 4-40s. Link + Proxy performance enhanced at Transport Layer (TCP) stallscan occur when mobile host is stationary. and the Application Layer, for benefit over the GPRS Single Packet flight time distributions for downlink 1000 packets; random delay higher than 4s + Aggressive Web Browser Pipelining over GPRS
between packets

Transport Layer Enhancement (TL-E)


+ Splits TCP connection transparently into two halves + On GPRS side, it avoids TCPs slow start phase and instead, uses a fixed value of the congestion window (cwnd) and clamps it for the full connection duration + TCP clamped cwnd value = optimistic value of the bandwidth delay product of the GPRS downlink + Share this window amongst all flows to the same mobile + Bandwidth: Variable and Fluctuating, Max: 4.15 KB/s downlink and 1.5KB/s uplink using a 3+1 channel GPRS phone host + ACK Compression: Link Layer (RLC) effect, both on uplink as + TCP cwnd clamping leads to: (i) Minimization of excess queuing, (ii) Faster startup for short flows and, well as downlink (iii) Quicker recovery from losses + Losses: Relatively rare during stationary conditions

TCP Problems over GPRS


+ Sluggish slow-start phase in TCP due to very high RTTs (so, short TCP flows have higher transfer times) + Excess queuing at wireless gateway during long flows, leading to RTTs of over 30s. This leads to gross unfairness to other existing flows and a high probability of timeouts on initial connection request for new flows + Slow recovery after timeouts, due to excess queuing. + Spurious TCP timeouts due to occasional link stalls

Application Layer Enhancement (AL-E)


+ Allows pipelined requests from pipeline capable browsers, even if there is no such support from servers

Experimental Setup and Results

Browser Behavior
+ Most Web browsers aggressive, open many concurrent TCP connections + Good for Wired-Internet (reduces download times), but has a high cost over GPRS + Cost: Signaling and Connection setup overhead + Also, may lead to saturation of downlink buffers + HTTP/1.0 very inefficient, HTTP/1.1 is better, however, what about pipelined connections? + Pipelining HTTP requests over GPRS can yield higher downlink utilization

Experimental Web download tests over GPRS show that: Use of a performance enhanced mobile proxy combined with a moderate support from a pipeline Web browser, reduces mean web page download times by about 15-20%. A browser making few concurrent connections but aggressively using pipelining on them can substantially improve response times over GPRS.

Open Issues and Future Work


1. How can web browsers be made to dynamically adapt to the underlying network heterogeneity? 2. How to minimize the possibility of Head-of-Line (HOL) blocking effects with pipelining? 3. How to maximize pipelining efficiency in presence of resource inter-dependencies with dynamic web content? Read On! http://www.cl.cam.ac.uk/users/rc277/gprs.html

You might also like