Facebook Inc. hosts an annual event called Systems @Scale at its Menlo Park, California campus that, as the name implies, is dedicated to the systems and technologies powering the world’s biggest data centers. The company uses the invite-only conference partly to showcase some of its own internal infrastructure innovations.

Today, data center experts once again converged on Menlo Park for Systems @Scale to see what Facebook has been working on. The company unveiled Tupperware, a software container management system for massive data center networks, and a storage technology called Delos that serves a niche but foundational role.

Tupperware

Tupperware runs on millions of Facebook servers worldwide and helps power practically all the company’s services. Its role is to orchestrate the software containers in which the social network deploys application code. Containers allow developers to bundle their applications into lightweight, portable packages that can easily move among different servers.

Tupperware makes full use of this portability. When a developer releases a new piece of software, the system packages it into a container and automatically deploys it in the appropriate part of the Facebook network. If the developer gives the instruction to set up the software, in say, Facebook’s Southeastern data center cluster, Tupperware will pick out the specific section it deems to be the most suitable.

The system spreads out containers in a way that minimizes the number of servers sitting idly without being used. It also takes hardware availability into account. If a machine is offline because of maintenance or a technical problem, Tupperware will relocate containers to a different node.

The software places a particular emphasis on accommodating stateful container applications, such as databases, that are tasked with storing important information for long periods of time. Tupperware allows such applications to “weigh in” on orchestration decisions to reduce the risk of outages. If an update is being released, for example, the affected workload can instruct Tupperware to delay the rollout until it’s confident that there won’t be any issues.

Delos

Delos, the second internal technology that Facebook debuted today, serves an entirely different function. It supports the company’s control plane services, the foundational software systems that manage its network infrastructure and dictate how data flows among workloads.

Like any complex application, a control plane service requires storage infrastructure to hold onto important information. And it needs a data management layer to interact with that infrastructure, which is the role Delos fills.

One standout feature of the technology is upgradability. According to Facebook, Delos can be modified and enhanced as the services running on top of it change over time without having to be taken offline, which spares users from having to deal with downtime.

Delos is also flexible. It can provide data to control plane services in just about any format they might require without introducing too much additional complexity to the environment, which eases maintenance.

Photo: Facebook