Did https://pypi.python.org/pypi/ became huge and slow?

[Chris Angelico]

On Wed, Mar 11, 2015 at 12:42 PM, Cameron Simpson <cs at zip.com.au> wrote:
> On 10Mar2015 23:27, Chris Angelico <rosuav at gmail.com> wrote:
>>
>> Ah, sure. Especially the bit about the cache :) Though I usually find
>> that download times on https vs http aren't all that different; it
>> takes a little bit longer to establish the connection, but less than
>> the variation in DNS timing. Most of the time, the cost of encryption
>> isn't the bottleneck - on a large download, transfer throughput is
>> usually what takes the time, and on a small download, it's too hard to
>> measure the cost anyway.
>
>
> Unless you're using a satellite. High packet latency (and DNS and cert
> validation via said latency) put a VERY noticable upfront cost to every
> HTTPS connection versus HTTP. Once underway, they're essentially the same
> (per connection).

Wouldn't DNS timing also be affected by that? Yes, you get more steps
in the handshake with HTTPS, which means the latency is that much more
significant; but all handshakes are going to be slow on satellite. And
all it takes is one dropped DNS response and you're back to the top of
the flowchart, hence my comment about the variation of DNS timing.

But in any case, my main point still stands, when we're talking about
a 9MB download. Encryption cost on the body of the transfer isn't
significant.

ChrisA