2013-01-13 05:10:13 +00:00
|
|
|
Whilst developing the [Radiopaedia iOS app][radiopaedia-app] I ran
|
|
|
|
|
into the problem of very slow download speeds within the app. The app
|
|
|
|
|
has packs of content available for purchase and download. The download
|
|
|
|
|
speed was fine in the simulator but on the device it was painfully
|
|
|
|
|
slow<sup>1</sup>. Both the Mac running the simulator and device were
|
|
|
|
|
connected to the same Wi-Fi network. Profiling the code didn't reveal
|
|
|
|
|
any issues either.
|
|
|
|
|
|
|
|
|
|
[radiopaedia-app]: http://radiopaedia.org/articles/radiopaedia-ios-radiology-app
|
|
|
|
|
|
|
|
|
|
Some searching revealed people complaining about YouTube videos
|
|
|
|
|
downloading faster over 3G than Wi-Fi. There were a number of theories
|
|
|
|
|
and proposed fixes, including [people seeing faster rates when
|
|
|
|
|
downgrading their access point to 802.11b][80211b]<sup>2</sup> but I
|
|
|
|
|
found one particular article (that I can no longer find) that said the
|
|
|
|
|
power management in the iOS Wi-Fi stack interfered with download speeds
|
|
|
|
|
when communicating to a high latency destination over a high speed Wi-Fi
|
|
|
|
|
connection. For example YouTube servers in the US from AU<sup>3</sup>.
|
|
|
|
|
|
|
|
|
|
[80211b]: http://web.archive.org/web/20120511020134/http://qelix.com/blog/2008/08/31/get-better-wifi-speeds-on-iphone-3g
|
|
|
|
|
|
|
|
|
|
With this in mind I tested hosting the content packs on an Australian
|
|
|
|
|
server, instead of in Amazon S3 accessed via CloudFront. This showed a huge
|
|
|
|
|
improvement in download speed. At the time CloudFront didn't
|
|
|
|
|
have an Australian presence so we switched to hosting the packs on
|
|
|
|
|
RackSpace's CloudFiles, which uses the much more extensive Akamai CDN
|
|
|
|
|
and ensured that users of the app would have the best possible download
|
|
|
|
|
speeds no matter where they were.
|
2013-01-13 02:42:46 +00:00
|
|
|
|
|
|
|
|
<div class="seperator"><hr class="left">✦<hr class="right"></div>
|
|
|
|
|
|
2013-01-13 05:10:13 +00:00
|
|
|
Fast forward a year or so and I started managing podcasts on my phone
|
|
|
|
|
instead of via iTunes. I noticed that some of the podcasts I was
|
|
|
|
|
subscribed to were horrendously slow to download. Some would take hours
|
|
|
|
|
to download, particularly ones hosted a long way away, such as in the
|
|
|
|
|
UK or The Netherlands. I also noted that podcasts that used a CDN
|
|
|
|
|
downloaded quickly.
|
2013-01-13 02:42:46 +00:00
|
|
|
|
|
|
|
|
This seemed to be the same problem I had encountered with the Radiopaedia app.
|
|
|
|
|
I wondered if telling the iPhone to use a proxy on the local
|
2013-01-13 05:10:13 +00:00
|
|
|
network would work around the issue.
|
2013-01-13 02:42:46 +00:00
|
|
|
|
2013-01-13 05:18:39 +00:00
|
|
|
To test my theory I set up [Squid][squid]<sup>4</sup> on the Mac mini
|
|
|
|
|
that's connected to my TV. I configured the iPhone to use the proxy and
|
|
|
|
|
compared the speed with and without the proxy in use. You can see this
|
|
|
|
|
in action in the screenshots below.
|
2013-01-13 02:42:46 +00:00
|
|
|
|
|
|
|
|
[squid]: http://www.squid-cache.org/
|
|
|
|
|
|
2013-01-13 05:18:39 +00:00
|
|
|
In the first screenshot the phone isn't using the proxy. I have let
|
|
|
|
|
it download a third of the file to be well clear of [TCP's slow
|
|
|
|
|
start][slow-start] and it is estimating 57 minutes to download the
|
|
|
|
|
remaining 176.1 Mb (53 kb/sec). The second screenshot (only a minute
|
|
|
|
|
later) shows the result after I paused the download, changed the proxy
|
|
|
|
|
settings and then again let the download warm up a little. The estimated
|
|
|
|
|
time to complete the remaining 143.8 Mb is now only 5 mins (491 kb/sec)
|
|
|
|
|
-- a huge improvement.
|
2013-01-13 02:42:46 +00:00
|
|
|
|
|
|
|
|
[slow-start]: https://en.wikipedia.org/wiki/Slow-start
|
|
|
|
|
|
2013-07-28 07:27:22 +00:00
|
|
|
<figure class="screenshot">
|
|
|
|
|
<img src="/images/2013/01/podcast-download-without-proxy.png" alt="Podcast download without proxy" />
|
2013-01-13 02:42:46 +00:00
|
|
|
<figcaption>Podcast download without proxy, 53 kb/sec</figcaption>
|
|
|
|
|
</figure>
|
|
|
|
|
|
2013-07-28 07:27:22 +00:00
|
|
|
<figure class="screenshot">
|
|
|
|
|
<img src="/images/2013/01/podcast-download-with-proxy.png" alt="Podcast download with proxy" />
|
2013-01-13 02:42:46 +00:00
|
|
|
<figcaption>Podcast download with proxy, 491 kb/sec</figcaption>
|
|
|
|
|
</figure>
|
|
|
|
|
|
2013-01-13 05:10:13 +00:00
|
|
|
After the successful experiment I left the phone using the proxy and configured
|
2013-01-13 02:42:46 +00:00
|
|
|
launchd to start squid on boot:
|
|
|
|
|
|
|
|
|
|
<script src="https://gist.github.com/4521952.js"></script>
|
|
|
|
|
|
2013-01-13 05:10:13 +00:00
|
|
|
----
|
2013-01-13 02:42:46 +00:00
|
|
|
|
2013-01-13 05:10:13 +00:00
|
|
|
<sup>1</sup> Tested on iPhone 3GS and iPhone 4S.
|
|
|
|
|
<sup>2</sup> Original article linked from [Apple support discussion](https://discussions.apple.com/message/9564410#9564410).
|
|
|
|
|
<sup>3</sup> It appears there's YouTube end-points in AU these days.
|
|
|
|
|
<sup>4</sup> I installed squid using [Homebrew](http://mxcl.github.com/homebrew/).
|
