-
Notifications
You must be signed in to change notification settings - Fork 531
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
A gap range is missing #4304
Comments
The code you reference is when we get an acknowledgment for an ACK frame we sent. Th expectation is the peer knows our up to date state for all packets up to the largest packet number we sent in that ACK frame. So, then we can 'forget' all packets ranges before this number. This is not perfect, as this comment right below states, because it's theoretically possible if some of the packets less than this were significantly delayed, but this should be super rare, and we might already have sent an ACK for them anyways. But I don't think that's the scenario you're describing anyways. So, I'm confused about what the problem really is. In your table above, you seem to be indicating the MsQuic server is only acknowledging packet 3, and not packets 0 & 1 too? Is that a correct understanding? Why wouldn't it acknowledge everything it had received? So, I'd appreciate additional information so I may better understand your problem. Thanks! |
You're right that's not what I was describing. Sorry for the confusion. I'll try to be clearer.
As the MsQuic server is receiving an ACK#9 that is also acknowledging the ACK packet number 10, I believe it deletes packet numbers up to 4 so in the ACK table, the packet numbers#0,1,3 are removed. This is because the ACK packet number 10 contains the largest acknowledgement number of 3, and in the code, it drops up to "LargestAckedPacketNumber + 1" so that's 4. At this point, the ACK packet number 12 sent by the MsQuic server only includes the largest acknowledgement of 9, a gap at 8, an ACK range of 5-7. Therefore, the packer number 4 sent by the other endpoint is not indicated as acknowledged or lost by the MsQuic server. What I suggest is that we should only drop the packet numbers in the ACK table up to "LargestAckedPacketNumber" where it would be 3 in my example. So, when the MsQuic server receives the ACK packet number 9, the number#0 and 1 are removed as usual but the packet number#3 remains in the ACK table. By doing this, the ACK packet number 12 can include the largest acknowledgement of 9, a gap at 8, an ACK range of 5-7, a gap of 4, and an ACK range of 3. Thank you. |
Describe the feature you'd like supported
An msquic peer may response an ACK without a certain gap in the situation where it experiences some packet loss.
Proposed solution
The example below shows what I think is happening with the current situation.
ACK frame (number#12), sent to the client, does not identify packet number#4 as a gap in its ACK ranges after receiving ACK#9. To tackle this, I have modified a source file in the core in this directory msquic/blob/main/src/core/ack_tracker.c.
In the code, the "LargestAckedPacketNumber + 1" removes an ACK range entirely, and because of this, it cannot represent a gap range for the subsequent numbers if they are missing. So, I have changed it to dropping all packet numbers that are less than the largest acknowledged packet number, which is from "LargestAckedPacketNumber + 1" to "LargestAckedPacketNumber".
Additional context
No response
The text was updated successfully, but these errors were encountered: