text stringlengths 0 1.32k |
|---|
**Adam Stacoviak:** I can see why you're against it. |
**James Snell:** Well, you know, it's... |
**Adam Stacoviak:** They could have just gone the other way and just shrunk it, instead of the same thing back and forth, but just shrink it. |
**James Snell:** It added a lot of complexity. |
**Adam Stacoviak:** What are the plus sides to this complexity? You're talking about the bad side, but what's the...? |
**James Snell:** \[23:47\] Performance. Using that socket much more efficiently. I was doing a peak load benchmark here the other day with just the development image of HTTP/2 in core. We're at a hundred thousand requests at a server, there was fifty concurrent clients going over eight threads... Just to throw a bunch ... |
The header compression, for example, we were able to save 96% of the header bytes, compared to HTTP/1. Actually, it's 96% fewer header bytes sent over the wire with a hundred thousand requests. That's massive savings. |
If we're looking at the platform as a service where people are paying for bandwidth, saving that much is significant. |
**Adam Stacoviak:** A lot of money. |
**James Snell:** Right. |
**Adam Stacoviak:** They'll spend that money in memory, though. |
**James Snell:** \[laughs\] Yeah, they'll make up for it in other ways. That increase in performance is significant, you can't discount it. With the fact that TLS is required, there is an improvement in security, but there are definite tradeoffs, and anyone looking to adopt HTTP/2 has to be aware of what those tradeoff... |
One simple example is the fact that the status message in HTTP/1 - you know how you have the preamble on a response, HTTP/1.1 200 OK - that OK doesn't exist in HTTP/2. They've completely removed the status message. So no more "404 Not Found." It's just "404." No more "500 Server Error", there's no "Server Error." |
**Adam Stacoviak:** Just the number? |
**James Snell:** Yeah. There's no standard way of conveying the status message. They just completely removed it from the protocol. Well, there are existing applications out there that use the status message, and actually put content there that the clients read. Now, it's not recommended, and HTTP/1 spec doesn't assign ... |
**Adam Stacoviak:** That's a bummer, because people will stop saying "200 OK" now, they'll just say "200." |
**James Snell:** They'll say "200", yeah. "404 Not Found", the whole jokes... Nobody will get it anymore. So if you look at Node's API, or things like Express, they have "Here's how you set the status message." Well, that's a breaking change in those APIs when you go to HTTP/2, so we have to make a decision of how clos... |
**Adam Stacoviak:** So it makes upgrading or changing to HTTP/2 a very deliberate choice. |
**James Snell:** \[27:55\] Yeah, it's gonna have to be very deliberate, and it's only gonna be in very simple scenarios, which probably aren't realistic that somebody would be able to say, "Okay, it works in both." It's gonna be a thing where you have to design your application specifically for HTTP/2 in order to take ... |
**Adam Stacoviak:** It's kind of putting a high barrier in front of it, too... I mean, you can't expect adoption of what is, as you said, a better performing protocol if you put a mountain in front of it. |
**James Snell:** Right, right. |
**Adam Stacoviak:** No one's gonna wanna climb that. It's less enjoyable, or less likely, or whatever. People do it... |
**James Snell:** We have lots of people that say they really want this. They really want HTTP/2, and we have a lot of people that are talking about it not necessarily for user-facing - putting up webesites anyone can access - they wanna put it in their data center, and have server-to-server communication be much more e... |
There's opportunities there where you don't have to necessarily worry about the TLS; you could do a plaintext connection and you'll get a far greater performance out of it. But again, it has to be a very deliberate choice. |
**Adam Stacoviak:** So HTTP/2 is this something that you're solely working on, or do you have a team working on it with you? |
**James Snell:** Right now it's been primarily myself. I'm working on growing that team of contributors. |
**Adam Stacoviak:** Is it in IBM or is it open source contributors. |
**James Snell:** It's open source. I'm doing everything out in the open on the GitHub repo... |
**Adam Stacoviak:** Is it on your user then? |
**James Snell:** We're doing it under the Node organization. So if you got at github.com/nodejs/http2, all the works being done there. |
**Adam Stacoviak:** I saw that repo there, but I saw Ryan Dahl in there, so this is not a new repo...? |
**James Snell:** No, it's a clone of the NodeCore. |
**Adam Stacoviak:** Okay, I understand. |
**James Snell:** Even though the decision hasn't been made to get it into Core yet... |
**Adam Stacoviak:** You're assuming it is... |
**James Snell:** We're assuming it is, and developing at this. |
**Adam Stacoviak:** I'm following you... I was wondering -- I expected it to be a module, but then again... |
**James Snell:** \[laughs\] It's being implemented in such a way that we could easily extract it out as a native module if we needed to, if that decision was made. |
**Adam Stacoviak:** With all this change, wouldn't it make sense just to cut the chord and... You know, one thing Thomas and Sam were talking about was verbally and documentation-wise deprecated; don't do anything to the way it responds, or using anything within the Core. Why not just verbally deprecate it and then...? |
**James Snell:** It's way too early for us to do that. HTTP/2 is a very immature protocol. It still has to be proven, and the vast majority of the web is still driven by HTTP/1. Going out there and saying, "Okay, we're gonna deprecate this" when HTTP/2 has not yet been proven would be very premature. |
**Adam Stacoviak:** So what do you do then - you just offer both? |
**James Snell:** Both, yeah. And just say that Node is gonna be a platform for HTTP development, 1 and 2. There will be a mechanism - it's built into the HTTP specification - that you can actually run HTTP/1 and HTTP/2 on the same port. You can have a server that will offer both, and the client negotiates which one the... |
The fact of the matter is we can't get rid of anything in Core. You see that in things like the recent buffer discussions whether we deprecate things... We can't get rid of things that are so critical to what the Node ecosystem is doing; even having a deprecation message in there is problematic. |
**Adam Stacoviak:** That would ruin things, yeah. |
**James Snell:** \[31:59\] And something so fundamental as HTTP/1 - I don't think we would ever get to a point where we would fully deprecate it. |
**Adam Stacoviak:** Yeah, I'll retract that deprecation statement and say it more like, instead... Because when we were having a discussion about the options of deprecating things, it was not to put it in where it was a response, but more so in documentation, where it was frowned upon; it wasn't forced. |
You're obviously so much more closer; I'm just outside, looking in, but I'm thinking, if it's so deliberate to choose it, wouldn't it make sense (or potentially make sense, and this will be a decision you all eventually make) to offer it as a module instead. That way, you can have a clean break when it is time to move ... |
**James Snell:** It's a legitimate question. That's actually one of the decisions the CTC has to make. I have an opinion on it, but unfortunately it's not all up to me. We have to listen to the folks, to Sam and Thomas, and the ecosystem, and figure out what is the right approach to take. We're not close enough yet to ... |
It would be a native module, and all the things that come along with native modules. There would be some considerations there, but if we needed to, we could. Like I said, I have my opinion on what it ultimately should do, but it's up to the community, it's up to the Core team to make that decision, for whatever reasons... |
**Adam Stacoviak:** Cool. Let's close with any closing thoughts you might have on this subject. Anything I might not have asked you that you're like, "I gotta put this out there before we close down." |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.