MsQuic – QUIC Implementation from Microsoft (opens in new tab)

Have you considered implementing any parts of this in F* (so they can be verified) and extracting back to C, as is being done for TLS?

https://project-everest.github.io/

Sorry, this might be a bit off topic but it's something I've been excited about for a while. From what I've heard Microsoft is sort of getting behind gRPC. Have you tested using QUIC as a transport layer for the gRPC client/server libraries that MS maintains? Are you seeing notable performance benefits with QUIC as the underlying channel?

Any strong reasons you chose C?

Maybe I skimmed too quickly but I didn't see this mentioned in that blog post.

Was there a requirement from other teams?

Do you think you've adquently built something that can be used as a transport layer first, compared to googles attempt that seems more like a http snowflake first and a transport layer second.

QUIC has the potential to be helpful in game development but suffers from an overly specialized approach.

Will BBR congestion control be supported? Without it QUIC performance cannot match that of TCP, in my environment at least.

Does QUIC or your implementation of it support application control of keying or is it always based on x.509 certs and CAs?

edit: the spec (https://tools.ietf.org/html/draft-ietf-quic-tls-27) seems agnostic on this. Also simple APIs especially in security are important, so supporting certs only is no flaw in my book, just curious about the edges of how the OS QUIC could be used in the future.

I know it's not your team, but now that it's going in Windows kernel, maybe you could ask around... Any plans to support QUIC for data transfers from/to Azure Blob Storage?

I recently had a project in my college course to implement MP-QUIC, but there was sever lack of resources on it. What do you guys think of MPQUIC vs QUIC?

Was't QUIC was standarized as HTTP/3 what makes it call QUIC not HTTP/3.

will it be possible to use the « sendfile » system call to do zero copy file transfers on a quick connection?

Will you add support for boringssl?

will this be integrated in some rpc framework like gRPC we all can use ?

is this using BBR congestion algorithm ?

I'm with you, and I'm happy that it seems like they are trying to become a good steward in the software industry. I am impressed by the quality of Windows 10, up to about a year ago I used Linux daily because it was a better experience than Windows 98 through 8.1, until recently. Lately, I find it's way easier to run Windows 10 with WSL rather than screw around with Linux trying to get things to work correctly that I just plug in to Windows.

I hope they start giving more control to us, though, because I'm tired of having to firewall block telemetry and forced to have Cortana installed or whatever other garbage. If Microsoft allowed me to install Windows like I do Debian, where I can pick my packages and leave out what I don't want, and they also allowed for replacement APIs, so I could swap explorer.exe for my own version for example, I'd never use Linux again. But that'll never happen, so I'll just use tools to block that stuff for now and hope the Linux experience catches up.

I'm with you on the licensing side.

But C? come on Microsoft.

People cheering on Microsoft embracing things. Ten years ago I wouldn't have believed it. Today it's not even surprising. /s

Likely for the same reason everyone tries making their own web browser, even though the other guys' are all gratis (and many of them libre): When something is a platform, you are either a landlord or a tenant.

Google and Apple have seen what happens to Microsoft's tenants, so they decided to be landlords. Microsoft knows how awful a landlord it had been (and after decades of landlord-only status, suffered abuse as a tenant at Google's gmail and youtube platforms), so it also tries to be a landlord in every way it can; It couldn't attract it's own tenants to Windows Phone, IE11 and its own Edge, so it has to offer subleases on Android, iOS and Blink(=Edgium) these days.

QUIC looks more "behind the scenes" as a platform right now, but building your own is a very cheap hedge against ceding complete control of what could become a potentially fundamental platform to your competitors. So everyone does that.

I'm no fan of Microsoft, and I believe that Microsoft has been "beaten to submission" rather than "left the dark side", so to speak. But regardless of the overall technical quality or moral/values one assigns to Microsoft - they are a smart, politically and business oriented and savvy corporation. This is a "staying relevant and in control" move.

Possibly the result of an industry scarred by the experience of depending on a TLS library that everyone thought was secure because “hey everyone is using it”. Now they want to make sure they understand and control critical security infrastructure.

Multiple implementations of a protocol are by no means a bad thing. The opposite is a problem - too few implementations means that the implementations dictate the spec and creating something compatible requires implementing bugs of those implementations too.

That said, I’d be surprised if we still had this many implementations 5 years from now. I think at least some of those will become unmaintained and instead use what Google/Microsoft/Mozilla/Cloudflare have developed.

We are currently testing HTTP/3 support in IIS/http.sys internally. Cannot comment on any external product release timelines.

You mean IIS? HTTP.SYS isn't a webserver.

There is currently no standardization for QUIC APIs. You can check out the MsQuic API here: https://github.com/microsoft/msquic/blob/master/docs/API.md

Connection migration is part of the current Internet-Drafts. The generalized support for multi-path (i.e. usage of more than one path at the same time) is postponed to a future version of the protocol. You can follow the standards work here: https://quicwg.org/

Just like whole Linux, *BSD, Windows network stack, all these drivers for embedded systems... smh

Cross-platform means it runs on more than one platform, and you mentioned two. Those two are conveniently fairly popular as well.

It's MIT licensed, you can port it instead of complaining.