Why Amadeus moved its core Master Pricer application to Google Cloud
- 26 April, 2019 08:37
The Spanish travel tech giant Amadeus has moved its core Master Pricer travel search application - which powers metasearch engines likes Expedia and Kayak and processes up to hundreds of thousands of transactions a second - to the Google Cloud Platform, in the first step of an ambitious cloud migration strategy.
Speaking to Computerworld UK at the Google Cloud Next conference in San Francisco last week, Denis Lacroix, SVP of core shared services R&D, explained that the decision to move such an important application to the cloud came through pure necessity.
"Back in 2017 we started to see a possibility that we might be running out of space in our data centre," he said. "Master Pricer is compute-intensive and uses a lot of machines, a lot of floor space, and is growing in traffic 30 percent every year, so pretty much exponential growth.
"We could see that within three to five years we would be out of space. So we had two choices: either we pour more concrete and extend the data centre, or, we actually move the workload to the cloud."
His colleague Dietmar Fauser, SVP of technology platforms and engineering, added that of all the Amadeus applications, Master Pricer comprises "modern distributed Linux applications with in-memory data management technology, so there is no Oracle database underneath, which is why they were relatively easy to be containerised and put on [Google Cloud]."
During the migration, Amadeus had engineers from both Google and Red Hat on site to help them get to grips with OpenShift and the container orchestration technology Kubernetes. "We adapted Kubernetes quite massively to our needs," Fauser added.
So within 18 months Amadeus had lifted and shifted Master Pricer to the Google Cloud Platform, but why did it opt for Google and not, say, Amazon Web Services (AWS)?
"First of all it's a non-exclusive deal; we intend to also partner with other cloud providers," Lacroix said. The reason they opted for Google Cloud in this instance is twofold: a historic collaboration with Google due to its feed for Google Flights and the vendor's expertise around Kubernetes.
"Kubernetes runs simply very well on Google, it comes from Google. So on AWS and Azure there is a little bit more attention needed to ensure that it's really smooth," Fauser added.
And why not go all in on Google Cloud in that case? "[There is] fear of lock-in and also just commercial negotiation 101 really, to keep everyone awake," Lacroix said.
Amadeus says the key benefit of this shift is of course cost savings, which are primarily achieved through the greater flexibility of cloud compared to running on premise.
"You really have to spend a bit of time to understand the economics of the platform you're working with, the levers you can use, how quickly you can scale back your footprint after a campaign run by an airline or Expedia, for example, and the faster you scale back the cheaper it gets and you can see that the day after," Lacroix said.
"Currently we actually size things up for peaks. These peaks happen around Black Friday, January and Easter, but in between you have hardware just sitting there, just doing nothing," he added.
Fauser added that cloud also forces his developers to be more "dollar aware" as they now get instant feedback as to the cost of any change.
That doesn't necessarily mean going to the cloud is cost neutral, Fauser added. "You have to go with relatively high volume commitments to get a competitive price... on Google the bill only from log activity is reaching close to a million [dollars] a year already," he said.
Another benefit is latency. "Some of our large customers are based in the US or Australia and Singapore. So when we serve them from our data centre we have latency up to half a second," Lacroix explained. By running Master Pricer in Google regions closer to these customers, Amadeus can quickly improve that search latency.
The last benefit is more indirect and helps with recruiting and retaining talent because "working on these type of things is so much more sexy to advanced engineers than working on mainframe", Fauser added.
In terms of the broader strategy, Amadeus is in the process of retiring all of its mainframes in a bid to go 100 percent cloud.
"The move is a major step forward in Amadeus’ vision for cloud-based architecture that is globally distributed across both the private and public cloud, driving more speed, responsiveness, and resilience in global operations," the company said in a press release.
"In addition, Amadeus is now progressively migrating its shopping applications to run on the public cloud across multiple regions. This enables Amadeus to scale its system capacity faster and on-demand, which is particularly key in moments of peak travel."
Next will be a case of moving other applications to the cloud, whichever vendor that ends up being with.
"The data that Master Pricer works with, airline fares data, is read-only data ... So that's like a fairly easy use case. The other extreme is if you have a super large Oracle cluster. You can move that to the cloud, but it requires a lot more work," Lacroix said.
"What is important now is to take on the engineering efforts to bring more complex applications, with more messy data management [to the cloud]," Fauser added.
"The real problem is the data management - it's not so much the compute tier. You get VMs everywhere, but running complex, synchronised Oracle operations is a different ballgame."