Scalability of nepenthes
I did some further examination on the scalability of nepenthes. The testbed is a 2.4GHz Pentium III system with 2 GB of physical memory and a 100 MB Ethernet NIC. The system is running on Debian with Linux kernel 2.6.12 and nepenthes 0.1.5 in default configuration. I tested this setup with an increasing number of IP addresses assigned to nepenthes - ranging from just 256 up to more than 32,000. I measured the number of established TCP connections and system load for a period of one hour. This measurement was repeated three times to cancel out statistical effects or burst in the network traffic. The results are plotted in the following two figures:
As you can see, the scalability is rather good: At the beginning, it is (nearly) linear. However, when the system load reaches 1, the system is occupied with I/O operations and thus the number of established connections decreases. With better hardware (especially processor and NIC), the scalability would be better. Since nepenthes can also be deployed in a distributed way, more flexibility is possible...
Average number of established TCP connections
Average system load
As you can see, the scalability is rather good: At the beginning, it is (nearly) linear. However, when the system load reaches 1, the system is occupied with I/O operations and thus the number of established connections decreases. With better hardware (especially processor and NIC), the scalability would be better. Since nepenthes can also be deployed in a distributed way, more flexibility is possible...
".
