Domain goes all in with AWS

Real estate listing site Domain has just completed the shift of its entire infrastructure into the public cloud with Amazon Web Services.

The company started playing around with AWS mid-last year, for functions like image serving, but last month celebrated moving 100 percent of domain.com.au into AWS's Sydney data centre.

Domain parent company Fairfax last year moved its infrastructure - including the Domain Group - out of its owned and operated Pyrmont data centre and into Equinix's Alexandria facility.

But after dipping its toes in AWS waters, Domain Group decided it wanted to dive in head first.

Selling the idea to the board was hassle-free - Fairfax board members had already grappled with security and data sovereignty issues in public cloud previously through the company's move to Google Apps.

Similarly, Domain is not an organisation that deals with PCI DSS information, so shifting its entire operations to the cloud didn't raise the same security concerns considered by other organisations. 

Including disaster recovery and staging, Domain shut down more than 200 servers in the old data centres.

According to chief technology officer Mark Cohen, the move - which took six months in total - was successful in part because Domain did the preparation work to ensure it made the most out of the transition.

Domain is arguably the Australian poster boy for DevOps. Technology director Paul McManus implemented the model - which groups IT operations and development together based on task - over the last few years to make shipping new changes and features much more faster and efficient.

It has resulted in huge growth to the number of changes Domain was able to push into production at any one time - two years ago Domain was releasing changes once every three weeks, while recently it shipped 70 releases in a week, Cohen said.

"Legacy companies with a history of running data centre and doing old-school IT, when they talk about going to the cloud they think they are going to lift into the cloud but still use all the same software, just in AWS," he said.

"But what our team really understood is you have to do all the work to get the continuous deployment and integration happening and be in a position to ship stuff daily."

Part of the improvement can be attributed to Domain's embracing of the as-a-service model not just for infrastructure, but for software and platforms as well.

It's running Elasticsearch for search, image serving out of S3, and all its servers are virtualised and running on AWS instances.

"You have to tap into software-as-a-service. The higher up the stack you go with AWS, it's not just cost-effective, it's efficient," Cohen said.

"We are starting to use a lot more of their products - we now use AWS' message queing system [for example]. We've embraced platform-as-a-service and software-as-a-service, not just infrastructure."

Developing on .NET - when most of AWS' infrastructure is built for Linux - didn't turn out to be too much of a problem either.

"We ran into a few challenges, but we've had AWS guys on site with us for a while now," Cohen said.

"The main issue is with] licence overheads. The big SQL Server license and all the various Windows servers we've got running - you have to manage license migration and be very conscious about it. Once you go into auto-scaling, you've got to know that you're covered by licensing."

Domain attacked that problem by adding a human into the automated process.

"You can automate scaling and go beyond what you're licensed to run, so we make sure licensing is covered by doing periodic audits," Cohen said.

As a result of the shift, out of hours support calls to the Domain IT team regarding problems with the website have fallen by 75 percent and stayed at that level over the last three months.

"The website is much more stable running on AWS," Cohen said.

"An out of hours support call has a cash cost to the company, and the cost of making people get out of bed or lose time with their family to deal with an issue that's not their fault. It's not a good thing.

"So [the 75 percent drop in call volumes] for me is one of the best stats."

Domain's technology director, Paul McManus, presented at the CIO Strategy Summit this week.