<GitHub54>
[smoltcp] whitequark commented on issue #44: > A proper implementation would involve actual routing tables, but I propose we adopt this or other limited solution as a stop-gap for now.... https://git.io/v5d1E
<sb0>
rjo, how many registers would be required to implement latency compensation with pipelines instead of tweaking timestamps?
<sb0>
rjo, and did we agree that user-defined latency compensation by just adding a configurable offset to each channel was useful?
<GitHub153>
[smoltcp] batonius commented on issue #44: @little-dude It's kinda messy PoC right now, so I haven't been pushing, here it is - https://github.com/batonius/netstack/tree/smolnetd . So far I've implemented the `ip:` scheme enough for `icmpd` and `udpd` to use it and for `dns` and `ping` to work on top of them. `tcpd` is useless for now because it expects the `ip:` scheme to calculate TCP checksum.... https://git.io/v5dH2
<rjo>
sb0: about 100 bits * 20 depth registers on average for each SAWG RTIO channel, 18000 FF per SAWG channel.
<rjo>
sb0: yes. just one offset would go a long way
<rjo>
sb0: good. i start loosing track of SED. for me reading through the discussions each time is becomming erroneous and time consuming. could we pack it up into a design doc?
<rjo>
whitequark, sb0: re the crystal for the si5324 on sayma and kasli: is there a special reason that frequency was selected or was it just the one from the kc705?
<rjo>
scratch that. they don't specify anything else...