Skip to content

Latest commit

 

History

History
36 lines (28 loc) · 1.49 KB

discussion.md

File metadata and controls

36 lines (28 loc) · 1.49 KB

Discussion

  • So this is only about encrypted traffic?

    • Any traffic is welcome, we just don’t try to base this on peeking beyond L3 info.
  • How do you know which packets are worth recovering?

    • Today we don’t. If more L3 marking were to become available, we’d use it.
  • How do you transport your measurement-related information?

    • Forward info: In encapsulation extension (e.g., with sequence number). 
Reverse info: The same way we transport the ACK channel. To be defined with encapsulation.
  • How to relay congestion for non-ECN-capable transports?

    • Right now, we are only addressing ECN-capable end-to-end flows. But the answer in the future will have to be: By dropping. Or, actually, not even requesting a retransmission when a congestion event would need to be relayed anyway.
  • How do you avoid spending more for LOOPS encapsulation than the performance enhancement is worth?

    • LOOPS will need some management that is weighing this (and doing the pair setup in the first place).
  • How can you be better than bespoke satellite PEPs?

    • We can't, but we may be able to improve other path segments on the same paths of satellites. We may also be good enough.
  • Why don't you do multipath, optimization across several LOOPs pairs on the path, involve end hosts?

    • Crawl before walk; all of these are potential topics once a first version of basic LOOPS is delivered. For now, it seems host-to-network operation is a high-priority next step.