Results

As in the DAS results (subsection 3.2.3) December 15, 1999 is selected as a typical day, despite the fact that the data are more varying during longer time frames, caused by recent tuning of the TEN-155 network. Therefore, the long term statistics (page ; subsection 2.2) are not yet valid here. However, the data of this day are still typical for a working day.

In the following paragraphs some results will be presented 1) in the form of HTML tables, 2) as plots derived from the Applet plot window. Both are fragments of the net performance Web pages. In these paragraphs the sites are labeled with the titles specified in table 5. For reproduction reasons all results are here also presented as gray values.

Figure 5 presents the HTML table of the minimum roundtrip values [ms] at December 15, 1999 between the UU-36 site and correspondingly figure 6 shows the percentage roundtrip packets lost. Figure 7 shows the table of the throughput values [Mbit/s] at that day for those connections. Note that in the headers of these tables the title ``>> Site'' (``<< Site'') stands for the connection: UU-36 to (from) the site entitled Site.

  

Figure: The HTML table with the minimum roundtrip values [ms] between the UU-36 site and the other sites. The ZELAS site is not listed here, because at that moment there were no data available. Show figure in real size.

  

Figure: The HTML table with the percentage roundtrip packets lost between the UU-36 site and the other sites. The ZELAS site is not listed here, because at that moment there were no data available. Show figure in real size.

  

Figure: The HTML table with the throughput values [Mbit/s] between the UU-36 site and the other sites. The ZELAS site is not listed here, because at that moment there were no data available. Show figure in real size.

To make the tendency in the throughput data displayed in figure 7 more clear, figure 8 displays the plot of the throughput data between the UU-36 site and the other sites with a 100 Mbit/s interface (SARA and ZAM) at December 15, 1999 and the day before.

  

Figure: Plot of the throughput data between the UU-36 site and the other sites with a 100 Mbit/s interface (SARA and ZAM) at December 15, 1999 and the day before. See table 5 for a description of the sites. Show figure in real size.

In the following two figures the results of the 100 Mbit/s interfaces are compared with the 10 Mbit/s interfaces. Figure 9 compares the performance differences between 100 - 100 and 100 - 10 connections by displaying the throughput data between the ZAM site (100 Mbit/s) and the UU-36 (100 Mbit/s) and FOM (10 Mbit/s) sites. Reversely, figure 10 compares 10 - 10 with 10 - 100 connections by presenting the throughput data between the IPP site (10 Mbit/s) and the FOM (10 Mbit/s) and UU-36 (100 Mbit/s) sites.

  

Figure: Compare 100 - 100 with 100 - 10 connections by displaying the throughput data between the ZAM site (100 Mbit/s) and the UU-36 (100 Mbit/s) and FOM (10 Mbit/s) sites. Show figure in real size.

  

Figure: Compare 10 - 10 with 10 - 100 connections by displaying the throughput data between the IPP site (10 Mbit/s) and the FOM (10 Mbit/s) and UU-36 (100 Mbit/s) sites. Show figure in real size.

From figure 5, displaying the minimum ping values from / to site UU-36 (see table 5) the following can be concluded:

• There is no significant difference in roundtrip times during the 24 hours the values are shown. This appears to be typical.
• Most roundtrip times are symmetric in both directions. However, the difference in roundtrip times for the connections UU-36 >> FOM and UU-36 << FOM is remarkable. A possible explanation may be a different route for both connections.
• The roundtrip time to / from the RUS site in Stuttgart is shorter than the roundtrip time to the ZAM and IPP sites in Jülich. Off course this is explained by the fact that the TEN-155 ATM link is between Amsterdam and Frankfurt. The DFN route between Frankfurt and Jülich goes along Köln, so this route probably is longer.

From figure 7, showing the throughput values for the connections to / from UU-36 for all sites, and from figure 8, displaying the plots from the throughput results for the connections to / from UU-36 for the sites with a 100 Mbit/s interface, there follows:

• During day time there is a considerable performance degradation for the connections from UU-36 to the German sites when the throughput values during the night are compared with the values obtained during the day.
• There is no performance lost during daytime for the Dutch connections between UU-36 and FOM.
• There is a bit performance lost for the connection UU-36 >> SARA, however, there is also considerable lost for SARA >> UU-36. This asymmetry is typical. The reason herefore is not clear, but the relatively limited processing power of the SARA site may have some influence.
• Also during daytime there is an asymmetry in the connections between UU-36 and SARA. This may be caused by the fact that there are more active processes at the UU-36 site, while there are few processes active at the SARA site. The processes at UU-36 are a.o. using the network. This implies that the maximum available throughput at UU-36 for one process is somewhat limited. Note that the performance Web pages showed a load of one for UU-36 and zero load for SARA.
• The performance degradation for the UU-36 - SARA connections occurs in the time frame 10:00 - 19:00, so this may be caused mainly by local office hours. However, for the UU-36 - ZAM connections this time frame is 10:00 - 02:00. This may indicate that also traffic to / from the US does play a role here.
• Without sites near the backbone of the TEN-155 network it is difficult to analyze at which part of the route the congestion did take place.

From figure 6, showing the percentages roundtrip packets lost for the connection from / to UU-36, the following can be concluded:

• The most roundtrip packets are lost at 12:00. This is in good agreement with the largest performance dip in the throughput values (figure 7).
• There # roundtrip packets lost from UU-36 to the German sites are larger than to the Dutch sites. This is not surprisingly because more routers are involved. However, a # packets lost larger than about 5 % resolution are quite considerable.
• There are no packets lost during the night. This suggests that the package lost is induced by congestion and not, for instance, by a bad configuration of a router.

From figure 9, which compares the performance difference between 100 - 100 and 100 - 10 connections by showing the throughput values between the ZAM site and the UU-36 and FOM sites, the following can be concluded:

• During daytime there is not much performance difference between the connections: the connection ZAM >> UU-36 performs slightly better than ZAM >> FOM. In the other direction the difference in performance is less profound, but the ZAM << FOM connection performs a fraction better.
• During periods with congestion at daytime the performance of the connections from ZAM are better than in the reverse direction. During nighttime without congestion there is no significant difference in the connections with UU-36. However, in contradiction to the previous item the connection from FOM performs better than in the reverse direction.

From figure 10, where the performance difference between 10 - 10 and 10 - 100 connections are compared using the throughput values between the IPP site and the FOM and UU-36 sites, there can be concluded:

• There is no significant difference between the 10 - 10 and 10 - 100 connections.
• There are more fluctuations in the throughput values for the connections from IPP to both sites than in the reverse direction.