Lukasa

The plan here is to investigate the network efficiency of HTTP/2.0 versus HTTP/1.1 by gently hitting Twitter's API. The above script does 5 HTTP requests to Twitter and gets the responses.

In my tests I'm showing an insane different, whereby HTTP/2.0 transfers 250 kB of data and HTTP/1.1 transfers 2.2 MB. This difference seems insane to me, so I'd like to work out why it's happening.

To run this, you'll need Python 3.3. Install the requirements listed in requirements.txt, and then go to Twitter's API page and set up an application. Get the values listed in the script for your application, fill them in, and run them, using Wireshark to check how much traffic was sent. Let me know what you find!

Lukasa

Keybase proof

I hereby claim:

• I am Lukasa on github.
• I am lukasa (https://keybase.io/lukasa) on keybase.
• I have a public key whose fingerprint is 90DC AE40 FEA7 4B14 9B70 662D F25F 2144 EEC1 373D

To claim this, I am signing this object:

Lukasa

import itertools

def blocks(size, total_sum):
    def ascending(iterable):
        last = 0
        for elem in iterable:
            if elem < last:
                return False
            last = elem
        return True

Lukasa

A Thankyou To Google

Dear Google,

I'd just like to thank you for disabusing me of the idea that JSON is under some requirement of well-formedness. I'd been labouring under the tyrannical misapprehension that some standards body had written down how JSON was supposed to work and that if I wanted to claim I was using JSON I would have to comply with that specification.

However, I today received a bug report indicating that a project to which I contribute was unable to make HTTP requests your Google Flights website could understand. I found this perplexing, but my eyes were opened when I saw your HTTP request. In particular, I enjoyed your freeform and inclusive definition of JSON. Allow me to show you what you've called JSON by showing you my SPDY request:

:host:www.google.fr

Lukasa

from requests.adapters import HTTPAdapter

class TimeoutAdapter(HTTPAdapter):
    def send(self, *args, **kwargs):
        kwargs.setdefault('timeout', 30)
        super(TimeoutAdapter, self).send(*args, **kwargs)
        

Lukasa

Start-Date: 2014-06-20  14:36:56
Commandline: apt-get install openssh-server
Install: ncurses-term:amd64 (5.9+20140118-1ubuntu1, automatic), python-urllib3:amd64 (1.7.1-1build1, automatic), openssh-server:amd64 (6.6p1-2ubuntu2), openssh-sftp-server:amd64 (6.6p1-2ubuntu2, automatic), ssh-import-id:amd64 (3.21-0ubuntu1, automatic), python-requests:amd64 (2.2.1-1, automatic), libck-connector0:amd64 (0.4.5-3.1ubuntu2, automatic), libwrap0:amd64 (7.6.q-25, automatic), tcpd:amd64 (7.6.q-25, automatic)
End-Date: 2014-06-20  14:37:01

Lukasa

Google's server is providing an incorrect pseudo-header field. Their response contains the pseudo-header field ':status' with the value '302 Found', in violation of the draft. The relevant section is Section 8.1.2.4 of the h2-14 draft, which reads:

A single ":status" header field is defined that carries the HTTP status code field (see [RFC7231], Section 6). This header field MUST be included in all responses, otherwise the response is malformed (Section 8.1.2.6).

HTTP/2 does not define a way to carry the version or reason phrase that is included in an HTTP/1.1 status line.

Lukasa

# Needed for boring connection logic
import ssl
import socket

# Needed for the cert work
from binascii import hexlify

from cryptography import x509
from cryptography.hazmat.backends import default_backend
from cryptography.hazmat.primitives import hashes

Lukasa

import collections


class HTTPHeaderMap(collections.MutableMapping):
    """
    A structure that contains HTTP headers.

    HTTP headers are a curious beast. At the surface level they look roughly
    like a name-value set, but in practice they have many variations that
    make them tricky:

Lukasa

import time

from hyper.common.headers import HTTPHeaderMap
from urllib3._collections import HTTPHeaderDict

def timeit(method):
    def timed(*args, **kw):
        ts = time.clock()
        for _ in range(10000):
            result = method(*args, **kw)