Pixelbliss - Fotolia
Ocado has been around for nearly 18 years. Over that time, it has gained a lot of experience in delivering groceries in the UK.
Its original business model has evolved to a point where it supports not only the company’s own online grocery fulfilment, but also a growing number of online grocers, who buy its platform as a service (PaaS) as part of a managed customer fulfilment service.
The Ocado Smart Platform (OSP) is a key part of the company’s expansion plans. In the UK, it runs the Ocado online grocery service as well as parts of Morrisons’ online grocery business, and Ocado recently signed a £183m deal with US retailer Kroger to use OSP. Other retailers include France’s Casino, Canada’s Sobeys and Sweden’s ICA Group.
Ocado describes OSP as an enabling platform provided to grocery retailers as a service. It gives access to Ocado’s physical infrastructure to run robot-enabled warehouse operations for the single pick of products, together with the entire end-to-end proprietary software applications required to operate an online grocery business – from user interfaces, through warehouse operating and control systems, to logistics, route planning and optimisation.
Traditionally, the software to run the warehouse operations was hosted in a dedicated datacentre.
“Our previous platform was monolithic and ran as an active-passive cluster,” says Chris Dabrowski, Ocado’s general manager for infrastructure and operations. “But we wanted to embrace a cloud native architecture, which required a horizontally distributed active-active cluster. This enables microservices to scale horizontally, allowing us to shift workloads between availability zones.”
Dabrowski says Ocado needed a platform that would allow it to manage multiple warehouses simultaneously. “This is a fairly large challenge and we wanted to make variability as low as possible,” he says.
The company also wanted to develop its platform in a way that would enable it to provide a manageable service, he adds. “This is why we chose a best-of-breed technology stack with Kubernetes, giving us the lowest cost of maintenance and the simplest way to automate.”
Describing how the platform is set up in a typical warehouse, Dabrowski says Ocado operates three availability zones, each using OpenStack for compute and a Ceph cluster for storage. On top of these, the platform runs Docker containers that are used to build out the microservices architecture.
Dabrowski says: “We can run warehouses in the UK and abroad for multiple customers and support multiple warehouses per customer. In the warehouses, we manage the technologies to enable customers to deliver groceries in their geographies.”
The service Ocado provides is delivered as a PaaS, says Dabrowski. The customer provides a warehouse, into which Ocado installs its robotic automation grid. Robot trolleys roam the warehouse on a grid above millions of grocery items, and can assemble a typical 50-item shopping basket order in five minutes. “We operate the grid on behalf of our OSP customers, and we do it at scale,” says Dabrowski. “If we lose an availability zone, the system rebalances.”
Read more about Ocado technology
- To reduce the cost of high-end datacentre servers, Ocado has developed a private cloud that runs on low-end PCs.
- Ocado’s head of data, Dan Nelson, reveals details of how the online grocer is using machine learning to bolster the responsiveness of its customer service teams.
One of the areas Ocado has been pioneering is the idea of building a datacentre as a mesh network of computing nodes. The open source project, called Kubermesh, is self-provisioning and self-healing and builds on Kubernetes. Within a few minutes of being powered up and connected, a new node is incorporated into the cluster. It will then be able to present itself, offering extra capacity, so containers can then be scheduled onto the new node as required.
In a blog post introducing Kubermesh, Ocado described how its largest grocery warehouse in Erith, which spans 563,000ft2, would need 400 nodes randomly dotted and wired together to create the mesh. The applications deployed on the nodes could then be strategically placed near other applications that they would often communicate with, for optimal speed and performance, it said.
Kubermesh effectively provides a bare metal, self-hosted, self-healing/provisioning, mesh network Kubernetes cluster.
Mike Bryant, platform automation engineer at Ocado, explains: “Kubernetes provides us with the software to deploy and run all our containers for microservices. A key advantage is its ability to self-heal, which makes it incredibly reliable. We run multiple copies across all critical services. As long as we don’t lose more than a third of our compute nodes, we can carry on running.”
Benefits of cloud native
For Dabrowski, cloud native is an IT paradigm shift compared with the way IT was run a decade ago. “People used to buy best-of-breed products like Oracle for their database and NetApps for storage,” he says. “They put too much on the same technology stack. If a product or service in the stack was not available for some reason, all the applications built on top of it were also unavailable, making repairs very difficult and upgrades costly.
“To upgrade, you had a huge monolithic stack to upgrade and the process took 18 months because you had to coordinate downtime. Cloud native offers a seamless migration, and upgrades need no intervention.”
A cloud-native architecture built using microservices running on a computational mesh comprises many on small blocks. If one or more fail or need updating, the overall service is still available, he says. “Maintenance, roll-outs and outages are all handled seamlessly,” adds Dabrowski.