Re: [quicwg/base-drafts] Handling of corrupt Retry packets (#3014)

[Nick Banks <[e-mail]>]

I'm not exactly what more you need than "this is expensive". Obviously the HW can't be built and then we decide what to do. I just did a quick and dirty perf test (not sure if RSS was completely configured), and these were some numbers of our software QUIC stack for a CPU limited sender:
```
Encryption          45.32%
UDP Send            36.86%
Stream Data Copy     6.07%
ACK Processing       3.65%
```
The leftovers are for various smaller things. This is also on a machine that doesn't support all the HW offloads we support in Windows, so the UDP number can go down (drastically, giving numbers closer to the one I gave before). Also, while I am no crypto expert, I do not believe Windows bcrypt APIs are overly costly, compared to other libraries.

I know UDP has been expensive in the past, but that's no longer the case. I feel we should not use that past behavior as an excuse to default to computationally expensive solutions, where other solutions are available.

And again, like mikkelfj said you can't directly compare HW and SW solutions very well.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/3014#issuecomment-535051316