why so much udp traffic fallback to tcp traffic on Android comparing to IOS

[Zhenliang Liao]

same version of cronet installed on both android and ios. plus the two types of clients speak to the same H3 server. 

why so much more tcp traffic fallbacked from udp on Android comparing to  IOS?

[Rajesh Mahindra]

Not sure what’s the scale of your setup, but we generally saw that android devices are typically used in poor network conditions. And we saw high correlation between poor networks and fallbacks to tcp. Hope that helps 

On Mon, Jul 4, 2022 at 21:26 'Zhenliang Liao' via QUIC Prototype Protocol Discussion group <proto...@chromium.org> wrote:

same version of cronet installed on both android and ios. plus the two types of clients speak to the same H3 server. 

why so much more tcp traffic fallbacked from udp on Android comparing to  IOS?

--
You received this message because you are subscribed to the Google Groups "QUIC Prototype Protocol Discussion group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to proto-quic+...@chromium.org.
To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/proto-quic/3b23aac8-2422-44b6-9bbb-40025f933ccbn%40chromium.org.

--

Take Care,

Rajesh Mahindra

[Zhenliang Liao]

" we saw high correlation between poor networks and fallbacks to tcp" ?? the more poor the networks is, the more tcp fallbacks?  how come? as we know, udp works better than tcp in most bad network cases. 

my question is actually concerning about os of client point. we saw android tcp fallback more than ios's 

[Ryan Hamilton]

The population of Android devices and iOS devices are not equivalent because of the different price points. As a result, I believe iOS devices tend to be on better networks (on average) than Android devices. I suspect this skew likely explains the difference you're seeing.

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/proto-quic/6ae787b1-5dd5-49e4-baff-f5d65229f74bn%40chromium.org.

[Rajesh Mahindra]

Yeah Ryan phased it better than I did, but yeah thats what we saw at scale. Typically, Android devices are lot more popular in countries such as Brazil, India, and that skews the network performance. And differentiating between bad network conditions and possible UDP blocking is a hard problem, so you ll likely see cronet switch to TCP in such scenarios.

[Ian Swett]

Chrome/Cronet's QUIC handshake has some fixed idle and total timeouts (4 and 10 seconds, respectively I believe), so on networks with lots of packet loss and/or large RTTs, the QUIC handshake can timeout prior to success.

But there may be other factors at play as well.

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/proto-quic/CAJ0ysq2BTARWQKVNkeF8_42vrV0B5dvOd3YqVc19pid8MQfyUg%40mail.gmail.com.