Shifting to AWS: a tale of two start-ups

When Sydney-based interactive TV technology company iPowow shifted from a local datacentre to Amazon Web Services three years ago, the migration was relatively straightforward.

A move from Red Hat Linux to Amazon’s Linux AMI was required, but that was managed internally as iPowow “learned as it went along”, said chief technology officer Dave Kierans.

For iPowow the driver of change was really bursty traffic generated from voting, interaction and offers made on live TV. Real time voting and other forms of interaction are also extremely sensitive to latency.

Managing that is a “fundamental design criteria”, Kierans told ZDNet at an AWS conference in Las Vegas.

For New Zealand-based weather forecaster PredictWind, however, a similar migration last year required “quite an investment”.

“The weather model we use is very complicated, with many pieces of software,” founder Jon Bilger said. “They are compiled to run on Linux, but not the version of Linux that AWS offers for the cloud.

“We had to recompile a lot of this code to make it run on the cloud.”

Bilger said the cloud is similar but has some differences to running a solution of physical hardware. 

“We had one developer working for six months full time to make the migration,” he said. 

IPowow's customers include broadcasters in Australia and more recently in the US, including giants such as ESPN and NBC.

The iPowow participation platform helps bring viewers and brands back together when viewers are second-screening. It also positions advertising and sponsor brands as an organic part of the programme, Kierans said.

AWS’s distributed datacentres allows processing to happen as close as possible to the customers.

The requirements are so specific and demanding, iPowow can’t use Amazon’s load balancing software, instead opting for open source Nginx, which Kierans said is used by 20 per cent of the high traffic websites in the world.

Kierans said with $2 million in new funding from Difference Capital iPowow is gearing up to service a strong American sales pipeline.

There are also new dimensions that can be added to the platform. Kierans said iPowow is eyeing social development to allow direct communications between the viewers using its apps.

where iPowow was managing sokies and latency, PredictWind went live on AWS in August last year searching for sheer computing grunt. 

The company had been buying servers from Dell and having them hosted by US-based provider LayeredTech. It had a server farm of around 25 machines before the AWS shift.

“The servers were three and a half years old, out of warranty and we were having some failures of RAM and hard drives,” said Bilger. “It was a real hassle to get the servers fixed and taking us away from our core business.”

PredictWind, which services boat owners and other weather watchers worldwide, was working its servers hard, running them for weather modelling 22 hours a day. The decision point, whether to buy new hardware or shift to the cloud, had arrived. 

“We’re glad we made the move.” Bilger said. “What used to take 22 hours a day to compute the forecast we now do in five hours. 

“We start up a lot of the most powerful Cluster Compute instances that AWS offers but only use them five hours a day.  So we are delivering our forecasts faster to our customers, which means better accuracy. “

Quadrupling PredictWind’s computing power also allowed it to extend its global forecast at 100km resolution from nine days to twelve days and increase its resolution to 50km. 

Better uptime and the and disaster recovery were bonuses along with the ability to scale for heavy loads, Bilger said. 

PredictWind uses Amazon’s EC2, RDS database and Simple Storage Service.