Hangout 25 Jan 2018
Reviewed minutes of last meeting
- tuexen@ discussed a quirk of the way the current FreeBSD TCP-FO server implementation works: accept() does not return until the three-way handshake completes or the server gets data in the SYN/ACK. This defeats TCP-FO's benefit for protocols (e.g. SSH) where the server sends the first byte. Should accept return right away, even if there is no data? tuexen@ will write to pkelsey@.
- jegg reports a case where cwnd is set to 0 (or apparently underflows) in some custom code. However, he also found a problem in cubic. It is computing the cwnd as a 64-bit value, then assigning it to a 32-bit value. So, for very large values of cwnd, it essentially assigns random values to cwnd. jegg will develop a patch to enforce a max of INT32_MAX. jegg will also audit the 64-bit calculations to see if they are correct.