Attendees: tuexen@, rscheff@, jtl@
Discussion
- tuexen discussed how to mitigate blackhole detection problems
- Action item:
tuexen@ has a review to ensure we don't do blackhole detection once we are at the minimum value D24308
- note: more complicated than needs to be in order to prepare for future changes
- tuexen@'s next fix will be to ensure you leave this mode on success -- not only with too many retransmission
- tuexen@'s plan:
- Allow the user to specify a length of time we will wait before trying blackhole detection.
- On the first retransmission, calculate the number of retransmissions are needed to hit the time the user has specified.
- Then, once the number of retransmissions has occurred, begin lowering the MSS.
- At each candidate MSS level, reset the retransmission count to 1 (in order to get quick retransmits) and try twice.
- The behavior on success will stay the same as today.
- Once we have retried twice at the lowest MSS level, we will revert to the previous MSS and reset the retransmission count to its previous value.
- Action item:
- Availability of Black Box logging tools: jtl@ will ask ocochard@ to upstream.
Discussion of D24237: rscheff@ will change both the in-order and out-of-order data path so we call sorwakeup_locked() [when necessary] after the SACK blocks are updated.
Discussion of D23281: rscheff@ is changing the default delayed ACK timer from 100ms to 40ms. We decided not to mark it for release notes, but to ensure the man page was updated to reflect the change.
Next meeting: 23 April @ 1400 UTC