slow start in a lab

asked 2019-10-03 11:36:35 +0000

nujo gravatar image

Hello All,

Could you please help me confirm to understand general TCP congestion mechanism in my lab. Testbed: mint (192.168.100.100) <> wanem <> win2k3 (172.16.0.100) I'm downloading a file from mint to win2k3 over http in two scenarios, packet capture is taken on win2k3: a. 10 ms delay set by wanem on a link to mint, not limited bandwidth, no packet loss. Resulting transfer speed 28Mbps. b. 10 ms delay, set by wanem on a link to mint, 1518kbps bandwidth, no packet loss. Resulting transfer speed 1.6Mbps.

Packet capture comparison shows much more bytes in flight in the not limited bandwidth file, so I understand this is because of much higher CWND, is this correct? I'm looking at the start of both captures for TCP congestion and fail to to recognize the slow start pattern. The fast capture shows more initial segments are transferred before waiting for ack. So how did the sender knew to limit transfer with CWND before even receiving DUPACK or RTO condition?

https://pasteboard.co/IAdeknC.png unlimited start

https://pasteboard.co/IAdeBiF.png 1518 start

https://pasteboard.co/IAdeYOk.png start compare

capture files https://filebin.net/bjcdj7dkempux5gb

edit retag flag offensive close merge delete