VPS vs CDN vs CDN with a domain name .... who will win? Will a CDN make you more $$$? (10)
VPS vs CDN vs CDN with a domain name .... who will win? Will a CDN make you more $$$?
Loading speed is considered a crucial factor when it comes to CVR of our mobile campaigns. There is no doubt that a fast loading landing page will beat a slow loading one, anytime of the day. But how big is the difference when we use an optimized VPS versus a CDN, do we relly need to put everything on a CDN or use VPS located in the same country as our target audience? Is there really a difference between <100 ms loading time and >500ms?
Let me find an answer for you 
I have used one long running PROFITABLE campaign for this analysis, here are the important details :
1. its an Asian GEO
2. its all mobile traffic, majority is carrier traffic
3. all is running through a LP so we can observe the CTR difference too
4. the data consists of 8 full days - I divided this into 4 period, 2 days each. I selected a period that also includes 1 day where the offer didnt convert well from whatever reason, to make it more "real" as we all know this happens all the time.
5.
Please dont ask me any more questions regarding the offer or traffic, I would like to keep the rest private, all the important data is already here
I have been using 3 LPs, all receiving the exact same amount of traffic, evenly split at all times running.
1st LP is hosted on a beyondhosting VPS in the USA- without CDN, the domain name is using dnsmadeeasy for faster lookup.
2nd LP is on rackspace CDN, without the use of a domain name.
3rd LP is on rackspace CDN again, this time using a domain name related to the target country, again using dnsmadeeasy.
Its exactly the same LP in all 3 cases, its a small and simple one, using about 30kb of graphics and some code, the overall size is around 40kb.
I used pingdom tools to measure the loading speed, unfortunately they dont have an asian point to test from, so I picked Australia as that one is pretty close to Asia. I ran multiple tests to get a range of values. The loading speeds were very different, the CDN LPs had 68-96 and 94-120, the one with a domain name came out worse, probably because of the extra time needed to lookup the domain name. The LP hosted on the VPS gave me mixed results between 600-800 ms.
However, when I tested the VPS LP from a point in USA or Europe, the loading speed went down to 240-490 ms, always below 500ms! So one way or another, the LPs on the CDN, were waaaay faster when it comes to loading times, but the VPS LP pretty much always loaded under 500ms too.
Now let me post the
Period1:
Pretty even results, the no-domain CDN wins by a bit. CND with domain is the worst.
Period2:
Almost identical results on all. CDN with domain worst again.
Period3:
This was the day when the offer didnt convert so well from some reason, as you can see all LPs are affected evenly. This time, the VPS wins by a bit.
Period4:
CVR is back, this time the CDN with domain wins.
All 8 days in one:
If you look at the total numbers, there is almost no difference at all. I was surprised to see the CDN with domain perform the worst, thou its just by a veeeery small bit. I was surprised to see the highest LP CTR for the VPS LP, didnt expect that.
So how to conclude this? There seems to be no need for a CDN setup in case you are already running on a fast VPS. Doesnt matter if its on the same continent as your target audience either. The key is to keep the loading speed below 500ms by using optimized graphics and as low code amount as possible.
I hope this helped someone
Interesting. One VERY important thing you didn't mention: what kind of traffic did you use ?
e.g. I would suspect the difference in CVR to be a lot lower on pops / redirect than on banner/MDSP traffic... because on banner traffic the guy clicked on a banner and is expecting something... which isn't the case of pops/redirect
Impressive CTRs matuloo, surprised everything was so tight!
What was the load time of your offer?
Originally Posted by wiifmdude
Originally Posted by wiifmdude
From this I would distill that, below a certain load speed, the user experience i.e. perception of "snappiness" is probably similar.
On their front-end the majority of the time-to-glass could actually be the browser rendering the DOM + carrier connectivity delays. Did you have any RUM on page e.g. like NewRelic Browser stats? The end user load speeds will be the real tell-tale here, server-based tests like Pingdom are pretty synthetic, but I suspect for real users they will all be similar with the network times and page download times being the minority component.
Keep up the good work!
Originally Posted by zeno
I put every static LP on a CDN anyways - its easy to setup, costs almost nothing ... so why not. In all my tests, CDN worked comparable or better than a VPS, so there is no risk in putting all LPs there.
BTW: I didnt have any external stats component on the page so cant really dig deeper into this.
Surprising! Almost no statistically significant differences there, I'd guess, on eyeballing the figures.
My initial thought is that as well as the rendering component, the carriers may just be pretty slow to everywhere. There's a big difference between a loading speed of 1 second and 200ms, but there's less of a difference between 8.2 seconds and 8.8 seconds. It would be fascinating to get some Real User Monitoring data on this - maybe a follow-up?
From my experience this changes a lot from country to country and types of traffic.
Also, as mentioned, being carrier traffic probably evens out all the results a bit, while having some high speed wifi or desktop traffic could do something difference.
Nowadays I just split test all the results until what works better for each country/traffic type/connection mix.
nice share man! Yah ive always had that issue with CDN's as well. Never could get them faster than a well optimize server with US traffic atleast.
Originally Posted by jasonaao
Home > Paid Traffic Sources > Mobile