Jumping from managed hardware to the cloud has paid off at Sydney-based travel agency Jetabroad, slashing deployment time and cutting costs, according to the company's chief technology officer, Benjamin Ranck.
The company, which is just shy of a decade old, is an online-only travel agent that specialises in international flight says Jetabroad's managing director, Darin Walters. "If you look at most places in the world, the bulk of airline tickets are sold online, in most mature markets anyway," he says.
"In Australia and New Zealand the bulk of domestic airline tickets are sold online, but if you look at long haul international air fares purchased by Australian and New Zealand customers, the vast bulk are purchased offline.
"So the opportunity that we see is that, over time, if you make the online user experience good enough, including the post-sale service, that opportunity should move online. In this part of the world it's a $5 billion-$7 billion opportunity, and the vast bulk of this is still offline."
Jetabroad MD Darin Walters.
Jetabroad sources tickets from multiple sources: In addition to the Amadeus, Travelport and Saber global distribution systems it sources tickets from non-traditional shopping tools.
"Every time someone performs a query on our sites, we have to pool all of those sources, crunch that information apply our logic, check multiple items and then get that information back to the customer as quickly as possible — which is a fairly significant data crunching exercise," Walters says.
Prior to a transition to Amazon Web Services' public cloud in September, Jetabroad was contending with scaling issues and seasonal fluctuations in online traffic, Ranck says.
Jetabroad CTO Benjamin Ranck
Jetabroad had managed servers that were hosted in a Tier 2 data centre in Canberra. "They did all the physical hardware infrastructure management but we did the management of virtualized OS-level stuff and networking to some degree," Ranck says.
"Jetabroad's had some pretty interesting growth spurts," the CTO says. "We had already run into a couple of scaling issues — there was a bit of firefighting and we could see on the horizon that there was this much bigger growth potentially coming down and we didn't want to be firefighting again. "
Cloud was seen as a way of dealing with traffic growth and a way of coping with the variations in that traffic in a cost-effective manner.
Shifting to the cloud for development as well as production has also meant a lot more flexibility, Ranck says: "When you're limited [by hardware] you end up shoehorning into servers that you shouldn't — you put a whole bunch of systems onto one server for testing, even though you know that in production they have to be on separate systems."
"You take shortcuts because you're limited by your capacity and once you've reached the envelope limit on a blade, it's really expensive to provision another blade when you only need 5 per cent more capacity, not 100 per cent more," he adds.
Jetabroad considered other cloud providers, including Microsoft's Azure service.
"At the time when we were considering [cloud] Azure didn't have the separate instance offering — it was still the Windows Azure platform, which would require a fair bit of architecture change for us," Ranck says.
That's not to say there was no rearchitecting involved in the shift to AWS, he notes. In particular previously manual processes had to be replaced with automation.
"The work we did was really about — how do we have 'configuration as code', so that literally none of our instances require some sysadmin to go through some sort of set of manual steps," he says.
"All of our instances now can be created at the 'push of a script'."
Shifting to the cloud has produced some intangible benefits that are hard to quantify, as well as some clear figures when it comes to cost and deployment time, the CTO says.
"There's a lot less wasting time, a lot less frustration, a lot more certainty in terms of being able to say 'Well this is my complete environment on this particular feature branch and I know that once I'm done here then, as long as it passes functionality tests, we're not going to have infrastructure issues'," Ranck says.
"A deployment used to take a day; now it takes an hour or two and that's mainly because we have to wait for sessions to drain out. Even that we're wanting to improve, until we get to the point where we can do a full production deploy within 15, 20 minutes.
"Before, that future was a pipedream and now we can really see it; we can say we can get there"
There were some teething issues, largely because it was very easy to spin up new instances. "We had a whole bunch of duplicate environments running, which became expensive — but we fixed that by setting some smart timers and so forth in place," Ranck says.
"So there were some teething issues in terms of process, but overall we've definitely saved money. Since we've done the migration we have increased traffic quite a lot and been able to still keep our costs below what they were before at lower levels of traffic. I think we still have some head room in there before we're even going to get back to [cost] parity."
Jetabroad is looking at expanding its use of cloud services, including potentially employing Redshift, Amazon's data-warehouse-as-a-service offering, to analyse the large amounts of data collected by the business.
"Currently I'm doing a lot of that ad hoc with my own workstations and so forth, but I'm hoping that once we get some of these analytical processes nailed down to where the business is happy, we could say 'Okay this is perfect for Redshift' and move it over," Ranck says.
"It's definitely got to be on the agenda," says Walters. "We've got all this data."
"We generate a tremendous amount of data and we have a tremendous amount of what I would believe is latent value in that data that we don't exploit at the moment and that's something I'd certainly like to address," Ranck explains.