GregZ
I'm seeing a strange discrepancy in packets per second between the two ciphers on a cpu with AES-NI. Openssl speed tests clearly indicate that aes-128-cbc is way faster than blowfish, however I get more packets per second out of 12B UDP frames with blowfish. Here is the setup:
SNDBUF=65536
RCVBUF=65536
TUN_MTU=48000
DEV="tap"
PROTO="udp"
I have to tunnels with a bond on top of each side. I'm using pktgen to generate 10 million packets on the bond interface and netsniff-ng on the other end to get a frame count. Here is what I see:
12B Packets
bf-cbc ~200k+ pps
aes-128-cbc ~120k pps
800B packets
bf-cbc ~640Mbps
aes-128-cbc ~840Mbps
Does anyone have any insight as to why blowfish is outperforming aes-128-cbc in packets per second with small datagram sizes?