Tablas de resumen de los datos recolectados u obtenidos de cada dimensión aplicada 56

All cases look at the performance of the entire email process, which includes both the upload and download stages as a single email transfer process. The performance is measured for this transfer process.

Case 1: The effects of changing end-to-end delay while keeping bandwidth and message size constant and how this affects time and throughput performance.

The email server processing time was measured by sending an email between a wired sender and receiver. The processing time was measured to be 4 seconds.

Uploads were done using TCP Reno and downloads were done using TCP Vegas. Time for upload and download stages were shorter for a shorter end-to-end delay. Therefore a shorter end-to-end delay improves email performance.

In the symmetric setup used by the simulations where the upload and download bandwidth is the same, it was found if one stage suffers from constant RTO then the whole email process is affected. The poor performing stage will dominate the time taken to transfer and cause the entire transfer process to perform poorly. In scenario two of the 1Mb message transfer the uploading stage causes the bulk of the time taken to transfer the message, and so affected the entire transfer performance.

Packets take a shorter time to propagate across a short delay network than a long delay network and resends for a damaged or lost packet will occur more quickly. A long delay is helped when TCP increases the window size which then negates the effects of a longer end-to-end delay. But it was still found that a shorter delay performed better. The simulations were done for all window sizes and a short end-to-end delay outperformed a long end-to-end delay. [Roccetti et al, 2005] found that it took longer to transfer data from a geographically further server which also had a higher end-to-end delay.

Table 6.7 the affects of changing end-to-end delay while keeping bandwidth and message size constant on time and throughput performance. The message size is 1MB (and 20KB) and bandwidth is 2048Kbps.

In real networks the uploading time will dominate the time to transfer a message as a mobile device transmits at a lower rate than it receives. This is because of design, function and power. Device manufacturers build the device to a higher receiving rate than transmitting rate. Also the operation of downloading is more common than uploading and finally to transmit at high data rates uses considerable more power than at low data rates [The Shosteck Group, 2001].

Table 6.7 The effects of changing end-to-end delay while keeping bandwidth and message size constant on time and throughput performance. The message size is 1MB (and 20KB) and bandwidth is 2048Kbps.

Message size

Scenario Upload time (s)

Server

processing (s)

Download time (s)

Through put KB/s

Total Time

20Kb

1 2.2 4 9.9 1.65 12.1

2 1.9 4 2.7 4.35 4.6

3 0.6 4 1.1 7.41 2.7

1Mb

1 46.1 4 438.2 2.05 288.3

2 4655.4 4 35.1 0.21 4694.5

3 31.2 4 33.0 14.66 68.2

Case 2: The effects of changing bandwidth while keeping end-to-end delay and message size constant and how this affects time and throughput performance. Illustrated in Figure 6.8.

Figure 6.8 The time to transfer an email file by each bandwidth for each message size.

An increase in bandwidth leads to an increase in performance. When the bandwidth increases, the time to transfer the email message decreases while the throughput increases. In the uploading stage bandwidths of 64Kbps and below experience constant RTO for message transfers of larger than 100KB causing the transfer to take very long. When only viewing bandwidths higher than 64Kbps a marked difference in performance can be seen, see Figure 6.8. The 384Kbps bandwidth has a throughput of 14Kbps while 144Kbps bandwidth has a throughput of 8Kbps for a 1Mb message and takes about half the time to transfer the message.

An increase in bandwidth results in better performing uploads and downloads, because TCP and the larger bandwidth together allow for a higher throughput. See Section 6.2.4 Case 2 for details. [UK Telematics online, no date] mentions that it will take the low GPRS bandwidth

0 1000 2000 3000 4000 5000

20KB 50KB 100KB 150KB 300KB 500KB 750KB 1MB

Email 2048Kbps Email 384Kbps Email 144Kbps Email 128Kbps Email 64Kbps Email 32Kbps

Time (s)

120 seconds to download a file that the high 3G bandwidth could do in 10 seconds; referring to the high difference in bandwidth and its effect on performance.

Case 3: The effects of changing message size while keeping end-to-end delay and bandwidth constant and how this affects time and throughput performance. Illustrated in Figure 6.9.

The larger the message size the higher the throughput achieved by the bandwidth.

Throughput increase is logarithmic with respect to increase in message size. All bandwidths perform similar for small messages. This is because the time to transfer a small message is short and TCP window size does not affect the transfer. As the message size increases the throughput increases until the maximum throughput for the channel bandwidth is reached and the throughput levels out. TCP controls this maximum window size based on bandwidth and channel conditions. Once throughput levels out, further increases in message size will yield insignificant increases in throughput. This maximum throughput is reached for low bandwidths at smaller message sizes than for higher bandwidths.

This indicates that it would be wasteful to use a large bandwidth to transfer a small message.

[Chakravorty et al, 2004(A)] discovered the same results that small files have similar performance over low and high bandwidths, whereas higher bandwidths achieve higher throughputs than lower bandwidths for large files sizes.

Figure 6.9 Transfer throughput achieved by each bandwidth.

The graph shows that for large bandwidths as the message size increases throughput continues to grow. The graph shape is logarithmic. The 128Kbps/144Kbps begin to taper after 500KB, while the larger bandwidths continue to grow. The 64Kbps and 32Kbps channels rapidly decrease in throughput as message size becomes larger than 50Kb due to constant RTO. [Dubois, 2005] discusses the occurrence of constant RTO and how increases in RTO can occur.

Based on simulation results and maximum users for a given bit rate displayed in Table 6.5 in Section .6.2.4 it is recommended that channel rate 64kbps is used for small emails while 128Kbps channel rate be used for large emails. Its performance is close to the higher data rates but more users will be able to enjoy the service simultaneously.