Early retransmit code

[siyengar]

If we're in the early retransmit mode, a loss_time is set. If the timer is not yet fired, however we send a packet when the timer is armed, the largest_sent_packet changes. When the timer does fire before this new packet is acked, largest_sent_packet != largest_acked_packet, so delay_until_lost == infinite and none of the packets will be declared lost, however we would instead set a TLP alarm.

Is this intentional? Seems a bit weird.

[martinthomson]

Note: Questions are better asked on the mailing list.

[siyengar]

Ya it was a bit fuzzy on whether this was a bug in the draft, so I raised this as an issue.

[siyengar]

I think this is not an issue any more

[siyengar]

sorry this is so old that i forgot why i raised this, will discuss with @ianswett offline and then close properly.

[siyengar]

More concrete example:

1. Send packet 1, 2, 3, 4

2. Ack 1,2,4.
   3 is lost
   largest_sent = 4
   largest_acked = 4

3. loss_time is set now to send_time(3) + 5/4 rtt

4. Send 5
   largest_sent = 5
   largest_acked = 4

5. timer for 3 fires
   DetectLostPackets(4) is invoked

since largest_acked != largest_sent

delay_until_lost = infinite

so packet 3 is not lost in this round and needs to go through another round

[ianswett]

@siyengar You volunteered to write a PR, can you try fixing this issue you filed?

[siyengar]

will do

[martinthomson]

@siyengar, some time has passed. Do you still intend to do this?

[siyengar]

@martinthomson put up #2021. Sorry about the delay

[janaiyengar]

I think this issue goes away with #1974.

…

On Mon, Nov 19, 2018 at 4:38 PM Subodh Iyengar ***@***.***> wrote: @martinthomson <https://github.com/martinthomson> put up #2021 <#2021>. Sorry about the delay — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#934 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AKjg1NfVp_OyMttHIYbz6nu0SG2GJ5Vfks5uw08WgaJpZM4QffCp> .

[siyengar]

@janaiyengar i think #1974 depends on how we add early retransmit timer back to that current PR. I think we might need something like this as well

[janaiyengar]

@siyengar, does the merged #1974 resolve this? I believe it does.