I am fortunate to have been involved with the Open Compute Project (OCP) since its inception in 2011. I was present at their first Networking Project meeting and have been part of some of its core contributions including ONIE and ONL.
In the past two years, OCP has made big strides in disaggregating networking hardware and software and creating a vibrant open networking ecosystem, enabling faster innovation, and creating better efficiency in data center deployments. Through the OCP vision, Facebook has taken a leading role in developing various open, disaggregated, and cost-effective hardware platforms, including a top-of-rack network hardware with Wedge 100.
These new hardware platforms have the potential of creating better efficiency in data center deployments. Yet for mainstream customers, there have been significant challenges in adopting OCP-based networks:
- The disaggregation of hardware and various software components provides great choice yet puts the burden on the end user to integrate the various components into an end-to-end system.
- While programmable interfaces are highly desirable, network engineers that are accustomed to Command Line Interfaces (CLIs) are suddenly required to transform into software developers, and understand the subtle differences between NetOS APIs and their semantics. We believe this is an unrealistic expectation.
- These devices are being inserted in traditional, non-OCP environments. Therefore, by adopting OCP-based devices, network engineers are required to deal with the complexity of multi-vendor environments.
The Apstra Operating System
AOS is a distributed operating system for the data center network that sits above vendor hardware — both servers and switches, including Wedge 100 — and delivers on three major goals:
- Operational turn-key simplicity and agility through powerful intent-driven automation.
- Operational control and visibility as network engineers are alerted to any anomalies that affect business delivery through real time telemetry.
- Massively reduced Total Cost of Ownership through powerful automation that enables hardware choice. Network Engineers are empowered to choose the hardware that best fits their business.
AOS enables the mainstream adoption of OCP-based networks by providing the following capabilities:
- AOS integrates all the components required to operate the network: configuration management through device specific drivers that utilize the appropriate RESTful APIs; high-resolution, closed-loop telemetry that takes advantage of the rich telemetry provided by OCP-based devices; and integrated device lifecycle management that leverages Zero Touch Provisioning (ZTP) and ONIE.
- AOS enables network engineers to operate their network as one system independent of the different device vendors in the network. As a result, AOS allows network operators to seamlessly integrate new OCP-based devices into networks based on traditional vendor equipment.
In short, AOS brings turn-key, powerful intent-driven, closed-loop automation to OCP-based networks and makes them a lot easier for mainstream customers to adopt them.
The Path Forward
Apstra has made a strong commitment to support OCP because we believe in the benefits of open collaboration, and in the power of choice. Integrating AOS with Wedge 100 is our starting point. We plan to support additional OCP networking and compute components to complement our portfolio of devices from traditional vendors.
Facebook has been blazing trails in data center compute and network architectures for many years. They have led cloud providers in articulating the importance of vendor-agnostic architectures, leaf-spine networks with vast amounts of east-west bandwidth, and homogeneous L3 designs. They have also worked with networking hardware vendors to help make the vendor’s systems programmable through APIs, and to provide access to high resolution real time telemetry.
By integrating AOS with OCP-based components, we are thrilled to bring highly agile and automated OCP-based networking to mainstream customers. AOS now supports both conventional devices from established vendors and OCP components, which provides customers a smooth, risk-free path to experiment with OCP-based deployments and take advantage of the massive operational efficiencies of Facebook-like network designs.