This page is part of the GENIUS project.

Quake 3 Aggregate Flow Traffic Results

Aggregate traffic flowing in and out of the server is important to observe as it affects network components near where the server is attached to the network, the network performance itself and Internet.

Date: 31/07/02 - 02/08/02

Inter-packet Arrival Results
Note: The following 4 graphs below display intervals and not packets. If one packet is delayed then two intervals may be affected - that is one with a higher interval value then next with a lower value. The legends on the right of each graph list the consecutive histograms set 2000 packets apart.
The Game Server sends updates to each of the clients every 50msec on average, where updates are N packets back-to-back, one packet to each active client.

That is, for N players you can observe one 50msec interval and (N-1) short, back to back intervals that are often well under 100 usec. Thus for 1 player we only see 50msec intervals, for 2 players we see a 50:50 split of both 50msec and shorter back to back intervals (<100usec), and so on. Generally speaking, (100*(N-1)/N)% of intervals will be less than or equal to 100usec and the remaining will be scattered closely around 50msec. The client interpacket arrival times will be around 50msec as can be seen here.

Figure 11.

Figure 12.

From Figures 11 and 12, it can be seen that the behaviour described above holds true for the majority of histograms. The remaining plots represent time periods where pkhisto's histogram covers part of a game and the number of players has changed. This occurs as pkhisto histograms are not synchronised to game start/stop times. For example in Figure 12, the magenta histogram (CIH 3) where approximately 28% of intervals were less than or equal to 100usec. Here 72% of the histogram had only one player present and then 2 players for the remaining time (verified with log files).

Figure 13.

Figure 14.

In Figures 13 and 14, the graphs show intervals between uncorrelated streams of packets converging on the server. Due to the uncorrelated nature the intervals range from less than or equal to 100usec up to some tens of milliseconds, with the upper limit dependent on the number of players.

Packet Length Results
Note: The following 4 graphs below display packet lengths. The legends on the right of each graph list the consecutive histograms set 2000 packets apart.
Figure 15.

Figure 16.

The length histogram of traffic flowing from the Server to Clients (Figure 15) illustrates an even distribution of packet lengths from 50 to 160 bytes, with a mean packet length of approximately 100 bytes. The cumulative packet length histogram (Figure 16) shows that the majority of packets have a length of 160 bytes or less.

Figure 17.

Figure 18.

For the traffic flowing from Client to Server, the packet length distribution is shown in Figure 17 and has a mean of 60 bytes. The cumulative packet length histograms also shows that the packets lengths are scattered about 60 bytes.

Packet Per Second and Data Rate Results
The following section refers to 4 games that were played over a 2 day period. The details of these games are listed in the table table.
Start Time Stop Time
Game 1 Wed 31 July 2002 14:28 Wed 31 July 2002 15:38
Game 2 Wed 31 July 2002 15:59 Wed 31 July 2002 16:47
Game 3 Thu 1 August 2002 17:44 Thu 1 August 2002 18:55
Game 4 Thu 1 August 2002 20:40 Thu 1 August 2002 21:51

Note: The following 10 graphs below display packet per second and data rate statistics.
Figure 19.

Figure 20.

Figure 21.

Figure 22.

For Server to Client traffic, both the packer per second(Figures 19 and 21) and data rates (Figures 20 and 22) clearly depend upon the number of players participating in the game. This value changes over time, as players join and quit a game. The rates recorded in the pair of packet per second and data rate graphs coincide with each other. It is also interesting to note that the data rates vary somewhat during the game, despite the number players.

Figure 23.

Figure 24.

Figure 25.

Figure 26.

In a similar fashion, the Client to Server packer per second (Figures 23 and 25) and data rates (Figures 24 and 26) clearly depend upon the number of players participating in the game. Although the graphs are similar to those for Server to Client, there is a fair degree more scattering between histograms.

Figure 27.

Figure 28.

The graphs above display the Average Data/Packet Per Second Rates for Server to Client and Client to Server traffic given the number of players (Figures 27 and 28 respectively).



 

Swinburne Homepage Site Map Search Index
 

Swinburne Copyright and disclaimer Privacy Feedback

Last Updated: Thursday 12-Dec-2002 14:54:22 EST
URL:
Maintained by: Grenville Armitage garmitage@groupwise.swin.edu.au
Authorised by: Grenville Armitage garmitage@swin.edu.au

IndexSearchSite MapSwinburne Home Page