How moving to Amazon Cloudfront increased our App Store conversion by 250% | Hacker News Reader

How moving to Amazon Cloudfront increased our App Store conversion by 250%

dev.anideo.com

39 pointsiamclovin14 comments

The only problem with this article is that it makes the assumption that app downloads are correlated with resource loading speed, which really could be completely incorrect.

The hacker news traffic was very targeted, watching a specific video for the content. Assuming the HN traffic was during period which saw 6-8% conversion, the real cause of the low conversion rate is almost certainly due to people clicking through from HN, watching the video and leaving without looking at the site at all. I don't even own an iPad, and frankly when someone links a video on HN or similar social sites, I watch the video and leave, regardless of the video sharing site.

I'd assume their more day-to-day traffic is getting there via other sources. Perhaps people saying "check out this app", perhaps people coming through search engines looking for ipad related video - essentially a much more targeted audience.

I could be completely wrong. Based on the information presented, I've made a huge number of assumptions. However, I'd be truly surprised if decreasing the load time of your static assets is really the driving factor in visitor conversions.

Edit: It seems their "App store downloads to web visit ratio" isn't really a conversion factor as much as a comparison of the two. My interpretation of it is that they now receive 28% of their web visits in downloads (for every hundred web visitors, they receive 28 app "downloads" (although it's not clear whether a "download" is a video load, or an app install)). Surely that's just because their web traffic has fallen after the HN spike? I'm not convinced that there's any real basis to the key claim made at the bottom.

Interesting datapoint, that the quicker loading of the page made that much a difference. Also I think the conversion ratio of 25% is really high. 1 in 4 getting your app? whow). Anyway, this pushed me to set-up a cloudfront distribution to do some tests. And while sometimes it's really fast, sometimes it is also a lot slower than our normal server (300ms vs 50ms or so).

Try other CDNs. If you're happy to pay, you can certainly get an awful lot better than Cloudfront.

Can you provide more info ? I'm interested in what CDNs do you think are better ?

MaxCDN (subsidiary of NetDNA). They're faster, the support tools are substantially better (invalidation on Cloudfront is the very definition of "suck"), the uptime is better, and the cost is a LOT lower.

Their support for .Net isn't very good, so it took me a while to get things working for their REST services. Overall a much better product with much better support.

CDNetworks, we tested global page load times with Gomez. CDNetworks was more stable and usually slightly faster.

FTR, Cloudfront is actually insanely expensive, even in comparison to "normally considered super expensive" alternatives such as Akamai.

(edit:) Huh... a few months ago Amazon dropped their bandwidth prices by 20%. These new prices are now more in line with "expensive CDN" as opposed to "more expensive than expensive CDNs". Still much more than I'm paying.

That said, I have never been quoted a "per-request" charge by a CDN, so I'm not certain how much more expensive that extra charge ends up making this service (although, of course, for large content that will amortize pretty quickly).

(Finally, Cloudfront has many fewer abilities than the "real CDNs" I've used, fewer POPs, and the story on "how long does my stuff stay in cache" is unclear. They certainly aren't on my radar yet, although they are slowly getting better.)

> and the story on "how long does my stuff stay in cache" is unclear

It sounds like you haven't really looked at CloudFront.

http://aws.amazon.com/cloudfront/faqs/#How_long_will_Amazon_...

No. If I download a few hundred 5GB files through Cloudfront and mark them by setting a Cache-Control header of 1 year, Amazon is not keeping that stuff at the edge. The rate at which things are prematurely purged is a key variable in differentiating CDNs. The fact that people (apparently, such as yourself, despite the dismissive tone you took in your comment) don't understand stuff like this is what Amazon counts on, and is why they are getting as many clients as they do.

I see people stick things on S3 and assume it's fast, because it's fast for them. But, the difference between cloudfront and S3 (US) from Asia is huge.

Out of interest, what is the difference for you of an S3 bucket in Asia versus Cloudfront?

The other thing I am never sure about with CDNs is whether they will actually keep your items resident always, even infrequently used ones. If they evict them and go back to origin it does not help. Has anyone done eviction tests on say cloudfront to see what the policies are.

10mb binary file

US Standard avg 112k/s took 1:31

Singapore avg 1290K/s took 0:07

Tokyo avg 895K/s took 0:11

Cloudfront avg 8758k/sec took 0:01

I'm in Hong Kong on a 100mb up/down line (and I can hit 10MB/sec up and down (at the same time)). The last 3 hops to cloudfront (ctinet is my provider):

  5  014136128014.ctinets.com (14.136.128.14)  4.673 ms  3.392 ms  3.210 ms
  6  119.27.63.37 (119.27.63.37)  3.624 ms  4.393 ms  3.298 ms
  7  server-216-137-55-171.hkg1.cloudfront.net (216.137.55.171)  3.519 ms  3.594 ms  3.340 ms

the relevant hops to singapore (it starts to get blocked at 12)

  7  * ge-1-0-3-xcr1.hkg.cw.net (203.169.57.33)  7.182 ms  4.902 ms
  8  xe-0-2-0-xcr1.sng.cw.net (195.2.10.30)  37.951 ms  38.858 ms *
  9  p38895.sgw.equinix.com (202.79.197.87)  41.037 ms *  43.288 ms

Right! I'm in Tokyo, with 4 different 100Mbps or 200Mbps fiber connections (different providers, at 3 locations). In Japan, that's normal btw, and costs about $50/mo.

Vanilla S3 is so slow, it's not any faster than a bargain VPS hosted in America. CloudFront is WAY FASTER.

In Asia, or at least Japan and South Korea, the slowness is psychologically compounded by the much higher speeds people are typically used to. So things hosted on CloudFront (or really any decent CDN with a presence in Japan) feel NORMAL, while most anything served from the USA (or Europe) feels SUPER SLOW.

At home or the office, I get 10MB or 20MB per second with ping times of 8ms or so. That's from either CloudFront or AWS Tokyo region. Or downloading Apple OS updates or OpenSolaris images, basically any company down with Akamai or similar.

With vanilla US S3 storage or AWS servers in US or US shared hosting like pair.com, it's more like 400KB/sec and 200ms+ pings. Same with Heroku (although I haven't tested that since last year).

I think my experiences in this regard are typical here, for both home and most business users. If the server is hosted abroad, it is ghetto-slow unless the owner takes appropriate measures (.jp server or CDN).

Not a priority for every site, I understand, but hopefully something worth considering.

Very misleading title. It should read: "How our slow css and js increased our bounce rate."

Reply on news.ycombinator.com