Transport Meeting 18 May 2017
gleb asks that people actually review D9770
- tuexen@ provided feedback; more encouraged.
tuexen@ wants to finish D9163
- bz@ provided feedback; tuexen@ waiting on confirmation that his concerns were addressed.
- jegg@ update about pmcstat when the system is busy. Will try some suggestions and report back.
- rstone@ still working on low-latency things.
- rstone@ saw a bug for stable/10 that was fixed by accident in stable/11 and higher. It is a corner case when you configure an interface in a space for which you already have a route and you take an interrupt at just the wrong moment. He will send a query to net@ to see if anyone cares.
- gallatin@ suggests that we spend time at the BSDCan Dev Summit transport working group talking about ways to provide backpressure information from the L2 drivers up to the higher layers. We need to make sure the "right people" come prepared to discuss that.
- glebius@ reports that he saw a performance regression with heavy TCP socket setup/teardown from earlier listen socket tests to current listen socket tests. The degradation appears in his baseline (stock FreeBSD-CURRENT), so it appears there is some sort of performance regression in FreeBSD-CURRENT. It will be difficult for him to do bisection. jtl@ will work with him to look at the commits in the intervening time for obvious code-change candidates. But, it might help if others can volunteer to help with the testing.
- We will meet in two weeks (June 2), even though it is the week before BSDCan. The focus will be on planning for the BSDCan working group session. Depending on how the working group session goes, we can look at cancelling the call the week following BSDCan.