We can encounter some figures like Fig. 1 when we are looking for the
relationship between Message size and throughput.
Fig. 1 It is quite easy to measure this relationship. The important result from Fig. 1 is, if the message size is small, the throughput is also small. This chatacterization can be attributed to the fact that if the message size is smaller, more calculation procedures are needed in general. ... I faced this problem these days. I want to get the traffic (not throughput :p) 155Mbps which is generated by just one machine, but the machine cannot generate such huge traffic due to its poor (the machine is powerful actually) ability. The machine's OS is FreeBSD 2.2.8, CPU is Pentium II 350, Memory size is 128MB. This machine is supposed to generate 155Mbps ATM traffic by using ENI-155 Card. There are two modes which this machine can use, IP over ATM and Native ATM. It was observed that Native ATM can generate larger traffic than IP over ATM. What I want to do is to generate nearly 155 Mbps traffic by this machine. However, there is a confinement. That is, I want to send every ATM cells seperately, not by group like sending large message by using AAL5. I tried Native ATM, AAL0. Here is the program that I made: ...actually the client program generates the traffic only from 6 sec to 59 sec. However, this has nothing to do with the point so ... 
Fig. 2 Fig. 2 is the result of the generated traffic. Green line corresponds to the traffic generated by the programs shown above. As a matter of fact, the traffic shown in Fig. 2 is not the real one, since the traffic is generated from 6sec to 59sec, but the traffic is the mean of the 1 minute. The actual traffic flow is, therefore, (60/53) = 1.1320755 times of the value shown in the figure. In this case, this will not cause a serious damage, so just leave it ... From the Fig. 2, the observed traffic flow is about 9.8Mbps, which was generated by using Pentium II 350 as a CPU. 
Fig. 3 Fig. 3 shows the amount of the traffic generated by Pentium III 500 machine. Only the CPU has replaced and other configurations are not be changed. The amount of the traffic is about 11.5 Mbps, which is 1.15 of the case of Pentium II 350. From Fig.2 and Fig.3, it can be concluded that in Pentium III 500's case, the ATM cells are generated only 1.15 times as much as in Pentium II 350's case. |