Ever wondered what the operational differences are between multi-cloud, hybrid-cloud and public cloud? Speakers at The Cloud Migration Summit - Russell Warman, Head of Infrastructure and Operations at Auto Trader UK, and Andy Nelson, Associate Director of Informatics & Cloud Operations at Illumina, as they speak from experience about the benefits, challenges, considerations and strategy in choosing their specific cloud architecture.
What would you say were your major drivers for migrating to the cloud? What did you want to achieve with the migration?
Russell Warman (RW): Organisational agility, make use of capabilities that would take us a while to get on premise was our primary driver.
Andy Nelson (AN): Our main driver was growth, the second was agility – the rate of change in DNA sequencing technology has been running at around 9x Moore’s Law for the past decade, and the only way to keep up with that is to continuously revise the data handling approach.
What factors did you consider when choosing your cloud architecture (hybrid, multi or public)? How did you assess your operational readiness?
RW: Our approach is multi-cloud, we use AWS, GCP and Azure to get the best from each provider. We assess operational readiness dependent on the features we are looking to use and the cloud provider’s maturity in that space.
AN: We started out trying to be cloud agnostic, with our platform being built to run either on AWS, or on open stack to allow on-premises operation – in hindsight this was a mistake, and since concentrating on a single provider we’ve significantly improved our cost, reliability and velocity.
Tell us about your current cloud architecture - do you have some on-premises data centres still in use? Which cloud provider(s) do you use and how did you evaluate which vendor(s) to use?
RW: Still predominately have on premise, with layer2 connectivity between all.
AN: Our strategy centres around single cloud solutions… the data transfer costs alone negate most arguments for multi-cloud or hybrid. This year we’re closing the last of our own customer facing datacentres having migrated those workloads into single cloud solutions.
How much of a concern was vendor lock-in for you when deciding on your cloud strategy? Was this a major driver or was it mainly driven by wanting to use the best-of breed tools from each public cloud?
RW: We use kubernetes as an abstraction for the majority of our compute, whilst we’re currently heavily into Google, that abstraction makes it feasible to move should we need to.
AN: Vendor lock-in was definitely a concern at first and attempting to mitigate perceived risk literally cost us millions of dollars… when we stopped trying to avoid lock-in and started concentrating on optimising to AWS, our cost efficiency and ability to execute significantly improved.
How did you decide which workloads and applications to keep on data centres and which to move to the public cloud as well as how you evaluate which cloud service to use for a workload or app?
RW: Our plan is to move all our applications. Either to the public cloud or use SaaS.
AN: At this point, we’ve decided not to keep any of our customer facing workloads in our own datacentres… the ability to iterate faster than hardware cost depreciation would allow, is just too valuable. The second half of that question, choosing the right services for a workload is one of the things I’ll be talking about – we’ve found that making the right choices here can have significant cost implications!
What was the biggest challenge you faced when migrating? How did you overcome this? Were there any specific tools or pieces of software used to achieve this?
RW: Dependency on complex on premise dependency. Latency sensitive applications when running multi-cloud. Organisational onboarding of the different ways in which we monitor and support the infrastructure.
AN: The biggest challenges were mindset, and habit. At first there was a lot of fear, uncertainty & doubt about cloud to be overcome, and even the most pro-cloud engineers would still fall into behaviour patterns that were optimized for a hardware world.
How much disruption did the migration cause? If you were to go back again, what would you have done differently?
RW: It hasn’t been that bad for us. The disruption caused is being offset by capabilities we’re unlocking as we go.
AN: With hindsight, going all-in on one cloud provider, and maximising the use of their services rather than building our own would have been the right approach from the start.
Has the migration been successful? Can you give some stats illustrating the benefits?
RW: So far, yes – it’s enabled us to increase the number of deployments per day (circa 350 per day on the new platform circa 65 per day on the old) and significantly reduce deployment times. release time is less than 2 mins now vs over 7 mins previously.
AN: Definitely successful – whilst we’ve expanded to multiple new geographic regions, and almost doubled the data volume annually for the past couple of years, our annual cloud cost has been almost flat (it actually went down slightly) during the same period… just handling that level growth in a self-managed DC would have taken more staff, expense, and time!
Meet these cloud experts at The Cloud Migration Summit on the 14th May 2019 in London. Andy Nelson will be talking about using public cloud in his talk “It’s A Party, Not A Lock-in…” and Russell Warman will give his presentation on multi-cloud strategy in cost-reduction and optimising the supply chain in his talk “The Road To Cloud”. Don’t miss out! Register now.