Vjom - Fotolia

How Google is riding the multi-cloud wave to win over the enterprise

The Google Cloud Next conference saw the internet giant share details about how enterprise appetite for multi-cloud deployments are working in its favour

One thing enterprises have never had in short supply when moving to the cloud is choice. Especially when deciding on where best to run their apps and workloads.

In the early days of cloud, many of these choices were presented to enterprises as being mutually exclusive. For instance, should organisations move to adopt a public or private cloud environment, and – if they opted for the former – should they entrust Amazon or Microsoft or Google with their data?

The perception that enterprises could only go one way or the other in cloud has largely disappeared now, as organisations have come to realise it is possible to combine different approaches to achieve the result they want. 

For proof of that, one only has to look at how the notion of operating a hybrid cloud environment has taken off in the enterprise sector in recent years, and the way organisations are becoming increasingly accustomed to sourcing services from multiple cloud providers.

The latter trend, in particular, is one internet giant Google claims to be actively reaping the benefits of, while analyst reports about how its share of the infrastructure-as-a-service (IaaS) market is growing certainly seem to bear this out.

The company expanded on this narrative over the course of its Google Cloud Next conference in San Francisco, between 8-10 March 2017, by getting customers to explain why a multi-cloud approach to sourcing IT services makes the best business sense for them.

A multi-cloud approach, in the case of many of Google’s customers who shared their cloud migration stories at the show, seems to mean combining GCP with what the Amazon Web Services (AWS) platform has to offer.

Read more about Google cloud

A common narrative amongst the customers who shared their stories was the fact they originally embarked on their journey to the cloud with AWS, but – often for cost and resiliency reasons – had seen fit to latterly tweak and expand their supplier strategies. 

That’s certainly how the move to multi-cloud played out for Waze, the Google-owned crowdsourced traffic data app, which shifted a sizeable portion of the infrastructure underpinning its service from AWS to GCP over the course of a couple of months.

Consumers depend on site reliability

Speaking at the event, Nir Tarcic, a site reliability engineer at Waze, said its turn-by-turn navigation service is relied on by consumers, as well as blue light service providers, to get real-time traffic information.

Underpinning the Waze platform is more than 100 microservices, and each one is hosted across multiple availability zones and datacentre regions for resiliency purposes.

“The most mission-critical ones are spread across multiple providers, Amazon and Google, so we can provide the best redundancy possible for our users,” he said.

Waze’s desire to source cloud services from more than one provider was hastened by an eight-hour period of DynamoDB service disruption AWS suffered in 2015.

“Our engineers made sure Waze did not crash on that day, and I’m happy to say it didn’t, but it came very close,” he said. “It was one of the triggers for us to realise we actually needed a multi-cloud solution, and can’t just rely on one provider.

“This was before our GCP migration. We wanted to spin up our GCP cluster sooner, but we couldn’t do it because weren’t ready.”

Prioritising the migration of more expensive workloads

The project was overseen by a team of just two people, he revealed, with the firm prioritising the migration of its more expensive workloads over to Google from the AWS cloud.

The company had previously relied on Amazon’s Elastic Compute Cloud (EC2) reserved instances to secure lower-cost compute capacity for its systems, but – after a period of prolonged use – the company encountered a number of issues with the platform.

For example, when using EC2, Waze would need to predict in advance how much capacity it would need over time.

But this can prove difficult to do, given the service often experienced huge, unpredictable surges in the number of people using it during rush hour or following a road traffic accident.

“Because all our systems use auto-scaling, that predicted number was never right. We either under or over estimated. It doesn’t matter how you look at it, we lost money,” he said.

“Reserving an instance means you’re locked into that one. Say, you have a new CPU [central processing unit] family being released and you want to use that. Sadly, you can’t because you’re locked in for one or three years. What can you do?”

Computer Weekly understands AWS has, over time, tweaked the way its Reserved Instances work in a bid to give users a greater degree of flexibility, allowing them to switch between instance types and availability zones, as per the needs of their applications.  

The benefits of using GCP

The cost and flexibility benefits of using GCP were touched upon by Urs Hölzle, senior vice-president for technology infrastructure at Google Cloud.

During one of the event’s keynotes, he claimed cloud companies that “force” users to pay upfront for years of compute capacity are no different to on-premise server manufacturers.

“Clouds are supposed to be elastic and free from capex or capacity planning, yet some cloud providers force you to pay upfront for the three years to get the best price,” he said. “How is that better than buying your own server for three years? There is no flexibility.

“Similarly, we’re forced to buy servers in fixed sizes, so we end up over-buying and still pay for it even though we can’t use it. We pay for a full hour even when your test run only runs for 10 minutes, so you pay for compute time you’re not even using, which adds up to 45% waste.”

This is one of the reasons why Google offers per-minute billing for cloud customers, while the introduction of its Automatic Sustained-Use Discount scheme in 2014 helps keep wasted cloud spend to a minimum, he said.

The latter, for example, offers discounts to Google Compute Engine (GCE) users who operate an instance for a “significant portion” of the billing month.

As such, users that run a GCE instance for more than 25% of the month automatically receive a discount for every additional minute they operate it for.

“As soon as you use a virtual machine for even a quarter of a month, you start seeing savings, and if you use a virtual machine for an entire month, you get a 30% discount. But you’re also free to stop,” he added.

Banking bigger discounts

At Google Cloud Next, the company also debuted the beta release of its Committed Use Discount scheme for GCE users, which allows them to bank even bigger discounts (of between 37-55%) when they sign up to a one or three-year purchase agreement.

Google’s Committed Use Discount scheme also gives users the option to switch between instance families and datacentre regions over the course of their agreement, which offers a degree of flexibility only a few types of AWS Reserved Instance users can lay claim to having.

According to a cost analysis of Google’s Committed Use Discounts versus Amazon’s Reserved Instances, carried out by cloud management company RightScale, using the former’s cloud services may work out cheaper on both a one-year and three-year basis, when certain usage conditions are met.

US-based Planet Labs is another AWS customer that has portioned off part of its infrastructure requirements to GCP to help it deal with the growing deluge of image data in its efforts to capture the changing face of the Earth’s surface generates on a day-to-day basis.

The organisation recently launched 88 satellites into space, as part of a project to start documenting the entirety of the Earth’s landmass on a daily basis, generating anywhere between seven and 10 terabytes of data in the process.

This data, along with an additional seven petabytes the organisation already has in its archives, is stored in GCP, which is also home to Planet Labs’ image processing applications too.

Flexible architecture and pricing models

During a customer Q&A session at Google Cloud Next, Troy Toman, director of engineering at Planet Labs, cited GCP’s more flexible architecture and pricing models as the main drivers for pursuing a multi-cloud strategy.

“Price isn’t the only reason, but that pricing model and their way of doing business with customers – and look we’re still big Amazon customers – but for this particular project it was a big factor for us,” he said.

To back this point, he talked about the amount of time and effort the organisation previously devoted to creating custom tools that could help Planet Labs get the best deal on Amazon EC2 Spot instances marketplace.

For the uninitiated, this is where AWS users can offer to buy – at a discounted rate – spare Amazon EC2 capacity, which in turn can help drive down the costs of running applications on AWS.

Because of the way GCP operates, the company has not had to create similar tooling since making the move over to Google, which frees him and his team up to work on other projects.

“It all comes down to how much you want to spend doing things that add value to your business,” said Toman. “For me, it’s a distraction from the work that needs to be done. So that was a factor in choosing Google.”

Deciding what to run where

Deciding what to run where was also influenced by a couple of other factors for Waze, said Tarcic. The most critical of these was whether or not moving a particular workload to GCP would make life harder for Waze’s developers. If so, it stayed put.

“We wanted to shield developers because we didn’t want to make their job harder just because it is easier for the company to run on GCP,” he said.

It was important, therefore, not to rush the spread of its workloads to GCP, so it had plenty of time to work out what should run where. It is an approach he would advise other multi-cloud-adopting organisations to take as well.

“A partial migration lets you mix and match provider solutions, and you can decide how to arrange the split between providers long-term,” said Tarcic. “You can enjoy the benefits of both worlds at the same time and experiment while you do so.”

Read more on Infrastructure-as-a-Service (IaaS)