Wednesday, December 06, 2017

Back to Sharing stuff I learned

I have not been regular in blog posts as I've just been focused on everything. I got lazy. Well, that is over.

At Shots Studios, a teen social network consisting of nearly 2M lines of code is no more. Shots is now a one-stop shop for select Creators. We are a Production Studio, Ad/Talent Agency, Talent Management Media company focused on creating timeless content. A 21st-century answer to getting great content from great creators in front of their audience.

Your internal monologue after reading this is how does this have anything to do with MySQL, HA, Scale, Coding; if not, this is still a good segway to explain how.

Shots the App, did really well yet not well enough to compete with Snapchat and Instagram. We did gain a lot of insight, mainly in what is called Influencers. A large percentage of time in growing the Shots platform was handling their cases of spikey scale. When Influencers posted they would promote their Selfie on other platforms sending waves of teens all at once to their data. Honestly, this was an amazing challenge to scale on a tight budget. Cold to Performant in millisecond time, with a 600% increase in load/concurrency suddenly. The short answer to scale this was to keep data in memory - From this, we understood that influencers reach and ability to move users is more effective than Display Ads. Period.

We did a huge analysis about our user base, and from that analysis, we made the decision to keep all "Influencers" in memory, and people who were the sticky users-the percentage of DAU that comes back with frequency. Next, to make sure that we did not saturate a network interface by keeping their data in memory on a single box, we replicated this subset of users among redundant pairs. Finally, we had to keep higher than normal frontends in reserve to handle the sudden burst without the startup delta of dynamic scaling pools.

Now we use a subset of the tech developed to mine, analyze, data about Creators. Creators, were influencers but now create, perform, direct, edit content thus they are called Creators. For instance, we use a custom performant event tracking system to monitor the social engagement of all creators. If you heard of a site called socialblade, I basically duplicated it at a much higher precision then their data.

With this we are able to tell which of a creator's content strikes a chord with users then we produce more of that performant content. For instance, With this insight, analysis, data collection and maximizing the reach channels on platforms like YouTube, Instagram with a shoestring budget we are making data-rich informed decisions.

No comments: