lookup. The hash table used linear probing which got very slow as the
hash table got full. Using std::map is about 10 times faster. The hash
table was slow enough that it took the majority of the time for me in an
ARP scan of a single address.
# nmap -sP 192.168.0.190
mac_prefix_init took 0.49261 s.
Nmap done: 1 IP address (1 host up) scanned in 0.59 seconds
# nmap -sP 192.168.0.190
mac_prefix_init took 0.04392 s.
Nmap done: 1 IP address (1 host up) scanned in 0.13 seconds
The memory usage of std::map is probably greater. The hash table used
19037 pointers and about 13000 structures of size 8 (on a 32-bit
architecture), or about 176 KB. Assuming the map has left, right, and
parent pointers, and a red-black indicator per node, the usage is 16
bytes per prefix plus 8 bytes for the structure data, or 304 KB total.
But this makes fingerdiff so much faster, I want to leave it in place at
least until this round of OS integration is done.
