Once the experiment is finished, TEM retrieves the log files to the host machine for analysis. A sample of the output format is presented in Listing 6.1.
lo: flags=73<UP,LOOPBACK,RUNNING> mtu 65536 inet 127.0.0.1 netmask 255.0.0.0
inet6 ::1 prefixlen 128 scopeid 0x10<host> loop txqueuelen 1000 (Local Loopback) RX packets 4 bytes 12406 (12.4 KB)
RX errors 0 dropped 0 overruns 0 frame 0 TX packets 4 bytes 12406 (12.4 KB)
TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0 sta3-wlan0: flags=4163<UP,BROADCAST,RUNNING,MULTICAST> mtu 1500 inet 10.0.0.4 netmask 255.0.0.0 broadcast 0.0.0.0
inet6 fe80::ff:fe00:300 prefixlen 64 scopeid 0x20<link> ether 02:00:00:00:03:00 txqueuelen 1000 (Ethernet) RX packets 437 bytes 241938 (241.9 KB)
RX errors 0 dropped 0 overruns 0 frame 0 TX packets 399 bytes 408311 (408.3 KB)
TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0
Listing 6.1:Sample output of netstat -iec collected per mobile node
A shell script filters out the data for the Wi-Fi interfaces, the IP address of the node and the number of bytes transmitted. This script also adds up the number of bytes transmitted for all nodes. This and the following paragraphs present the expected outcome as a result of the analysis of the data collected for this experiment. Figure 6.3 presents the results of plotting the overall number of bytes per second for JNFD using the 10k files (red) versus HTTP using the 10k files (purple) as well as JNFD using the 100k files (green) versus HTTP using the 100k files (light blue). The experiment with the larger files was run so as to ensure that the traffic generated exceeds the size of the Content Store on the nodes.
As expected, what is immediately striking about the diagram is that the overall network traffic generated for transferring the data is quite high. For transferring 30 times 100kBytes, or 3MB, the network nodes send 30MB worth of packets in the HTTP case. This is due to a number of
0 10,000,000 20,000,000 30,000,000 0 50 100 150 200 250 300 350 400 450 500 550 600 650 700 750 800 850 Time in seconds Accum ulativ e o v er all n umber of b
ytes in the netw
or
k
JNFD 10KB 50 Requests JNFD 100KB 30 Requests OLSR 100KB 30 Requests OLSR 10KB 50 Requests
Figure 6.3:Overall aggregated network traffic JNFD vs HTTP
factors: first, as data communication in a MANET is hop-by-hop and I am measuring the total size of packets sent by each node, the aggregate is influenced by how many hops a data packet needs to make to reach its destination. The overall traffic for the JNFD case is lower but still many times the size of the workload.
Considering the case of JNFD, I analyse Figure 6.3 in two parts: the first one points to the area where the traffic is higher in JNFD than HTTP, and the second part where the traffic of JNFD has a flat tendency while the HTTP traffic keeps rising. The first part is generated due to the traffic generated by intermediate forwarders that are searching for the content requested by the consumer. This traffic is high at the beginning due to the forwarders broadcasting the interest until they reach the producer, and due to the data packets being sent to the consumer using the reverse path. However, as JNFD caches the content, the next request of already cached content is served by the forwarders and not necessarily by the producer. Furthermore, as the effective path to content is shorter and as nodes have information in the FIBs, the number of interest broadcasts will be reduced. Consequently, the network traffic after the content is cached in intermediate forwarders is reflected in the curve flattening out.
Additionally, and from the consumer point of view, this experiment shows that the time required to deliver the content varies according to where it is coming from. On the one hand, it corresponds to the time utilised when the content comes from the producer. On the other hand, it corresponds to the time utilised when the content comes from the cache of nearby forwarders. For example,
the time required to retrieve a 100KB file from the producer is in average 2773.167 milliseconds, while in the case that the same file is retrieved from the cache of forwarders, this time is reduced to 7.975 milliseconds, in average. This last case highlights the benefits of provide caching to forwarders in order to reduce latency.
Another benefit of the use of caching in JNFD is reflected in the level of completion rate of the requested content. As a example, retrieving 50 times a randomly selected file of 100KB size from a producer, has a completion rate of 71%, in JNFD, while in OLSR this value is 67%. This supports the idea that content distributed across forwarders offers more availability of the information, and less dependency of the producer.
In the case of OLSR, the network traffic rises linearly. For both file sizes, the network traffic is constantly increasing. A degree of flattening of the curve could be expected if OLSR was benefiting from using established routes after a period of operation, dealing only with incremental changes to the topology. However, this does not seem to be the case, which raises the question whether the overhead of OLSR routing traffic decreases over time if there is no other traffic and if the nodes in the MANET are static. The following paragraphs present the result of OLSR network traffic generated in a MANET where nodes are following GPS traces, which is the current behaviour, and where nodes are standing at fixed locations and where no workload traffic is being generated.