The purpose was to output Json in a human readable form, making it easier to see what data is being transferred through one of my projects.

PHP does not yet have its own implementation of Pretty Printing inside its Json system yet (as of PHP 5.3.x), however there is some intriguing constants inside the source code (look for JSON_PRETTY_PRINT, which can now also be found in the json_encode manual page).

Meanwhile Zend Framework seems to have this solved through its Zend_Json class. With Zend Framework being open source you can easily get hold of the source code to translate it to any project and language you need, as long as your environment supports Perl compatible regular expressions (preg_split).

<code>
print Zend_Json::prettyPrint($json);
</code>

It is not quite perfect but it does the job.

I currently have an exporter storing full user accounts in a Json object saved as a file, part of this data is a column in the database which stores Json itself, it seems the Zend_Json::prettyPrint function will also go ahead and pretty this up also, not quite expected but cool none the less!

I’m sure there are many people out there with the same question!

The problem:

We want to use multiple ad networks to determine which provides the most engaged user, and which networks we are making our money back on. However Apple iOS App Store and Google Marketplace all make this difficult.

1. Installs are all through a single source (AppStore).
2. The AppStore swallows all source information.
3. Apple doesn’t allow us to put tracking codes into the install process.
4. Applications and the Web Browser are sandboxed. We cannot set a cookie and read it from within an application.

The solution:

I appreciate there are many sources of inaccuracy with this approach, however I think it’s a good start. We will need:

• Ad click tracking and forward to app store.
• Consume a click, recording a new user against the click with as much accuracy as possible.

This seems simple in concept but can be tricky to implement.

Ad Click Tracking

This involves writing a webservice which will record as many details about an ad click as possible, all open and available information given to us from the Safari browser will do.

Ad Service Notification

In our Game Servers we add a little extra to the Create New User functions, providing a callback with all of the details available which the browser normally can provide. This could even be done through a webview sent to a specific Game Server endpoint which can hopefully provide the identical information provided before.

Final Piece of the Puzzle

Consumption: We need to consume ad clicks as soon as we can. The longer we leave it the more inaccurate the results. Match up the first Ad Click with the first matching New User and attach the user to the click.

This should give us a picture of how successful and valuable our marketing campaigns have been when compared against each other.

Accuracy

Clearly there are massive sources of inaccuracy! I do wish Apple would allow us to put an arbitrary string into the install process ( think: appstore.com/myapp?source=my_advert2 ) as this would be far more accurate! But what other choice do we have?

If you want to find out more about the solution and how well it is working for us, please contact me and ask.

I found it quite a pleasant experience, until we launched.

My opinions follow…

Cheap

These are certainly cheap! Cheaper than many other reputable hosting options out there (offering root level access on dedicated Virtual Machines). The next level up machine only has half the processing power and is 32 bit only for about 4 times the price!

Fast

They’re certainly quick! 2 compute units of raw power! Provided you don’t have noisy neighbours… They held up well in a loaded situation! So whats the problem here then? Notice I mentioned noisy neighbours… When others want resources you will have them taken away from you without warning. It seems the micro’s don’t have a lower guaranteed processing limit which would have been nice! I’ve had complete instances fall over and become unresponsive for minutes at a time (up to 20 mins on one occasion). Rebooting will not help you here, as it takes time for it to respond to the reboot, especially when it doesn’t have time to even answer a ping! 12,000ms ping time anyone? on a gigabit (shared) network as well.

64 Bit

Don’t be fooled. 64 bit is great! But the next level up machine is only 32 bit. You cannot boot any AMIs built in 64 bit micro’s in a 32 bit m1.small. This caused me a problem as I had to rebuild my instances in a hurry in the middle of an expensive marketing push!

Shared

Clearly the problem is over-subscription. This is mostly evident in the way the micros are mentioned in the documentation. It simply states “up to 2 EC2 compute units burst”. What they fail to tell you is there is no minimum. If others on the shared physical host start to have a party, you’re drowned out in all the noise.

Conclusion

Don’t touch the t1.micros. They will run well in a development environment but they’re very bad news for production!

Though to break that rule I’m thinking of moving my current VPS requirements to Amazon, I don’t want to pay for an m1.small just yet so I think I’ll take the risk with a t1.micro. Still undecided though.

I’ve heard a rumour Amazon might be removing 32 bit only instance types from their portfolio, this would be very nice! Only time will tell..

I’ve recently been working on some pretty interesting projects, some which have been very successful at their design goals and have become the cornerstone for an entire company providing plenty of extremely useful information for people to draw interesting conclusions from.

I’m very proud of the stuff I’ve managed to create, not to mention it has allowed me to play amongst some very interesting players including providing solutions for the BBC and Electronic Arts.

Stay tuned for more as time goes on. I’m sure to let slip a few secrets every now and then…