Performance Evaluation and Discussion

Chapter 4

Name-based Message Routing for Home Network

Chapter 4

Name-based Message Routing for Home Network

0 0.005 0.01 0.015 0.02 0.025 0.03

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Numbe r of Node s / Numbe r of Locations

0 0.2 0.4 0.6 0.8 1 1.2 PDRC

EDRC SRC

Misdirections Per Hop for delivered messages

Figure 4-9 Average misdirections per hop for delivered messages

Vs Number of nodes/General location descriptions

0 0.2 0.4 0.6 0.8 1

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Numbe r of Node s / Number of Locations

PDRC - Cache lim it=64 EDRC - Cache lim it=64 SRC - Cache lim it=64 CBRP - TDb Size=64 CBRP - TDb Size=128 CBRP - Unlim ited TDb Size CBR - SubsTable lim it=128 CBR - SubsTable lim it=64

Undeliverd messages per sent messages

Figure 4-10 Average undelivered messages per sent messages Vs

Number of nodes/General location descriptions

Chapter 4

Name-based Message Routing for Home Network

one cache. But EDRC divides the cache size equally among the directions but the number of entries in each direction may differ. If a large number of entries have to be inserted into a smaller directional RC, some entries may be lost due to insufficient space while other directional RCs may have only fewer entries and more free space. The directions that may have lost some of its entries cause the messages to fail to reach the intended destinations.

The EDRC and PDRC show negligibly low message misdirections per hop compared with SRC, as these methods are able to determine the next hop. Even here, EDRC (Maximum is 0.0042 at 100 nodes) shows a slightly higher amount of misdirected messages per hop compared with PDRC (Maximum is 0.0028 at 100 nodes). The proposed PDRC takes advantage of the “free space” and the size of the directional RC is determined by the number of entries in each direction. This ensures that RC space is allocated fairly among the directions.

Routing in CBRP and CBR will not result in any misdirection/overhead (Here CBRP will not use source routing) as data messages are sent only after the TDb for CBRP and SubsTable for CBR is populated during network initiation. This allows CBRP and CBR to forward data messages without additional route discovery.

CBRP uses the TDb which enables the discovery of exact next hop for the known destinations. Also CBR uses the SubsTable at each node to achieve the same. But Fig. 4-10 shows that the delivery failures for CBRP and CBR are very high compared with the proposed methods with equal resources. In Fig. 4-10 PDRC shows a minimum of 99.2% delivery success rate at 100 nodes at cache limit of 64 bytes, whereas this value for CBRP and CBR are only 6% and 28.9% respectively (The success rate described here assumes there is no loss of packets due to wireless channel properties, and that loss occurs only when routes are not found). CBRP was unable to achieve 100% success rate even with unlimited TDb size. However given 256 bytes limit for SubsTable in CBR, it achieved 100% success rate. The proposed methods were also able to achieve the same at 128 bytes of cache space for the current network environment. Incase of a delivery failure for the proposed methods due to insufficient memory, source routing can easily be applied as an alternative

Chapter 4

Name-based Message Routing for Home Network

with the existing conceptual data structures.

Figure 4-11 shows the message overhead of the proposed methods and CBRP to achieve a 100% success rate (Here CBRP will use source routing if destination is not found even in TDb which causes the message overhead for CBRP). CBR will not cause any routing overhead as the SubsTable is already populated by all nodes in the simulation. However CBR requires larger memory size than other methods discussed here. Message overhead observed for the proposed methods are caused by the misdirection of data messages.

Figure 4-12 shows the power consumption of the proposed routing protocols as well as CBRP and CBR to achieve a 100% message delivery success rate. Here CBRP only showed very little variation in the power consumption when the memory size used for TDb was varied as it is mostly affected by the larger number of route setup messages generated by CBRP. Even though CBR does not cause any message routing overhead, it requires larger memory size to maintain the SubsTable.

The larger memory size is the main factor for CBR’s observed power consumption.

SRC shows higher power consumption than EDRC/PDRC due to its higher number of misdirected messages. Figure 4-12 shows that the all proposed methods on average consume less power when compared with CBR and CBRP. When compar-ing the three RC usages and the performance against CBRP and CBR, PDRC shows the best performance for any number of nodes considered in the simulation.

Fig. 4-7 and Fig. 4-8 were used to determine the most suitable value forℜ. Additionally Fig. 4-7 and Fig. 4-8 also represent the limitation of the use of the RCs.

It shows that after the network accumulates a certain number of nodes (i.e. the number of general location descriptions also increases) the misdirections per hop and delivery failures show a significant increase, for the used network environment settings. This occurs when the chosen RC (of size 64) reaches its threshold number of names. Here, the RC is unable to accept more names and maintain the value of ℜand thus it rejects the new names leading to the observed increase in the message misdirections and delivery failures.

Chapter 4

Name-based Message Routing for Home Network

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Numbe r of Node s /Numbe r of Locations

PDRC-128 EDRC-128

SRC-128 CBRP-128

CBRP-Unlimited CBRP-64 10^-3

10^-4

10^-5 10^-2 10^-1 10⁰ 10¹

Overhead Per Hop for Delivered Messages

Figure 4-11 Overhead messages per hop for delivered messages

Vs Number of Nodes/ Number of Locations

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Numbe r of Node s/ Numbe r of Locations

Power Units

PDRC-128 EDRC-128

SRC-128 CBR-256

CBRP-128 CBRP-Unlim ited CBRP-64

0 2܉10⁸ 4܉10⁸ 6܉10⁸ 8܉10⁸

Figure 4-12 Power consumption to achieve 100% success Vs

Number of Nodes/ Number of Locations

Chapter 4

Name-based Message Routing for Home Network

Figure 4-13. Sequential storage of entries in RC

A comparison was also performed between the RC usage and sequential storage of entries. The randomly generated names were stored sequentially in the RC of the same size. The entries were inserted until the cache was unable to store and maintain at least 30% of each entry (Fig. 4-13). The results showed that number of sequential entries stored in the cache of the same size is 51.8% lower on average than that of RC values showed in Fig. 4-4. During the comparison when there are entries already in RC matching a candidate entry more than 50%; the candidate entry is not inserted to RC as described in Algorithm_4-1. This is similar to aggregating the two entries in to one RC entry. The results showed that by allowing aggregation the number of keywords stored in the RC increased by 6.11% on average than the analytical values showed in Fig. 4-4.