HTTP 2.0 Draft – SPDY Protocol

http://tools.ietf.org/html/draft-ietf-httpbis-http2-00

by ejpastorino • 14 years ago

68 27 14 years ago

tools.ietf.org

ck22• 14 years ago

http://www.guypo.com/technical/not-as-spdy-as-you-thought/

The results show SPDY, on average, is only about 4.5% faster than plain HTTPS, and is in fact about 3.4% slower than unencrypted HTTP. This means SPDY doesn’t make a material difference for page load times, and more specifically does not offset the price of switching to SSL.

bcoates• 14 years ago

His benchmark indicates that SPDY doesn't magically make bad websites fast, which is to be expected. SPDY only makes sites that are sufficiently optimized that SSL's added latency becomes a limiting factor on performance. It also lets you not make some painful optimizations because you can multiplex more.

ricardobeat• 14 years ago

That test was done by proxying through a third-party. I don't see how that's any more credible than the other results from Google he cites and dismisses at the top. The restriction of having SPDY disabled for third-party domains also taints the results. "It won't make the web faster because not everyone will use it" is a silly argument.

He also dismisses the benefits of having encryption by default on every connection, that itself is worth a 3% slowdown.

dsl3• 14 years ago

He didn't just proxy through a random "third-party," he sent everything via Contendo. If anything that sped the whole thing up unfairly. Contendo had (before Akamai killed them) a killer DSA product that basically sucked a webpage in via the closest Contendo datacenter then transported it over their optimized and compressed backbone to the Contendo datacenter closest to the user.

SPDY hits you a bit harder in setup costs to be able to sling lots of requests back and forth faster. This is great for someone like Google who serves 100% of the on page content themselves. Anyone with advertising, third party content, or using a CDN for delivery might want to do some extensive real world testing before bothering to implement.

torrently2• 14 years ago

That absolutely horrible test is somehow posted to every story about SPDY. It uses a nonsensical methodology and adds nothing to the discussion.

ck22• 14 years ago

link to independent study that shows otherwise?

dsl3• 14 years ago

This guy is the chief architect at Akamai and probably knows a thing or two more about HTTP in general than most HN regulars. But please feel free to improve upon his work by proposing a better methodology and your results.

stereo• 14 years ago

Notice it says "on average". SPDY isn't a switch you flip on to make things faster. Very often, there are low-hanging fruits for performance improvement, especially in SSL: https://insouciant.org/tech/ssl-performance-case-study

fourstar• 14 years ago

If you're using Chrome, type this into your address bar:

chrome://net-internals/#spdy

ericmoritz• 14 years ago

oh man, this one is and its "clear host cache" will come in handing in the future. chrome://net-internals/#dns

mtgx• 14 years ago

Is the SPDY protocol done? Have they submitted it for the HTTP 2.0 standard, or will it remain separate?

wmf• 14 years ago

The point of this document is that SPDY will be the basis for HTTP 2.0. Until HTTP 2.0 is finished, I would expect Google and others to keep shipping SPDY implementations.

beatgammit2• 14 years ago

FWIW, Opera 12.1 and Firefox 15+ support SPDY, so it's not a Google-only protocol. The only major browsers left are Safari and IE, which are both slow at adopting new tech.

Source:

http://caniuse.com/#search=spdy

jaytaylor• 14 years ago

Bring it. We're eagerly waiting for Heroku to release SPDY compatibility.

cryptbe• 14 years ago

I'm so surprised that the security consideration section has no reference to CRIME [1], which has made Google and Mozilla to turn off SPDY's header compression in Chrome and Firefox.

[1] https://docs.google.com/presentation/d/11eBmGiHbYcHR9gL5nDyZ...

njs12345• 14 years ago

I think this version is basically SPDY 3. SPDY 4 (draft spec: https://github.com/grmocg/SPDY-Specification/blob/gh-pages/d...) uses a completely new header compression algorithm which is not susceptible to the attack used in CRIME.

jaequery• 14 years ago

there should be a disclaimer that you will only see performance gains if you make a lot of HTTP connections on your site. people expect some magic happening once they install SPDY, which is just not the case. but if your site is heavily on SSL, then you have nothing to lose by installing SPDY, but not much to gain either.

scott_karana• 14 years ago

While I appreciate SPDY, what's the point of a draft if adoption is so far not widespread?

Google's documentation seemed quite sufficient to implement, and until we have a proper reference implementation, this seems a little half-cocked.

Maybe I misunderstand the point of the IETF draft?

joshmoz• 14 years ago

Firefox, Chrome, and Opera support SPDY. Apache, nginx, and Node.js support it. Large Google sites support SPDY (gmail), so does Twitter, and Facebook is working on it (http://lists.w3.org/Archives/Public/ietf-http-wg/2012JulSep/...).

I don't know what qualifies as widespread to you, but this is at least a great start. And there are plenty of good implementations to reference.

Ironlink• 14 years ago

Also, CloudFlare: http://blog.cloudflare.com/introducing-spdy

jimktrains23• 14 years ago

I guess this is my naïveté showing:

1) What is the advantage of SPDY over HTTP/1.1? (Besides a possible (and debated?) speed up? (Due to sending assets w/o a request for them?))

2) Is it as easy to debug with tools like curl, wget, and telnet?

mattparlane• 14 years ago

1) A few things, notably server push.

2) curl and wget will probably eventually add support. telnet not so much, but then again you can't really debug HTTPS or even HTTP/1.1 (with chunked encoding) with telnet.

jimktrains23• 14 years ago

telnet is useful for somethings, but it does have it's limits. Often I'll compose my message and then paste it in (If I want to do something specific or have a header dump from chrome or something and don't want to convert it to a curl command).

openssl s_client is useful for debugging SSL issues. It gives you just about all the info you could want about the certs and then gives you an open pipe to the server that I treat like telnet above.

(Not really trying to argue. As things grow up more tools will support them and engineers like me will learn to adapt and debug them as they do the current generation of protocols.)

jimktrains23• 14 years ago

What's the difference between server push in SPDY and a multipart message in HTTP?

HTTP 2.0 Draft – SPDY Protocol

Related Stories