ygina

Going to simplify `commit()` and clarify the language surrounding replicas to address the remaining comments. It seems like sharding followed the convention where None meant 1 shard. For replication, i...

@fintelia I don't have any more changes to make. Is there a case you're concerned about in particular? It looks like the old code removes non-reader nodes in reverse topological...

In these experiments, I also increased the MTU from 1200 to 1460 in quiche to be comparable to TCP.

> Do you mean you changed curl? Yup I added a line in curl [here](https://github.com/curl/curl/blob/fc8ad0b23c9fd922ebdb89348d467df037da1775/lib/vquic/curl_quiche.c#L856) `quiche_config_set_max_send_udp_payload_size(qs->cfg, 1460)`. > can you describe your test environment a little more? * Built with...

> I guess you also need to update what @LPardue pointed in curl in addition to your change, because that code will limit the outgoing packet size, regardless of what...

> 10MB sounds like the example flow-control window size. What values does the server have configured, possibly defaults based on your nginx conf I'm not immediately sure what you mean...

It's looking a little better! At least, there is no sudden dropoff at 10MB. I'm guessing because otherwise, if we exceed the per-connection flow control limit the server spends resources...

I came across [this fastly article](https://fastly.com/blog/measuring-quic-vs-tcp-computational-efficiency) describing the tricks they use to get quic's performance on par with tcp/tls. Wondering if it's necessary to say, implement the Delayed Ack extension,...

@LPardue If I understand the parameter correctly, setting `http3_initial_max_data` to 100m should be good in terms of space for this single connection setup while not needing to increase the per-connection...

Cool! Here's the qlog from a 20M transfer using curl: [curl_20M.zip](https://github.com/cloudflare/quiche/files/10563712/curl_20M.zip). I'm currently checking out the visualization.