Datacenter Trends

AI-Driven Solution Tracks Packets Through the Datacenter

• By John K. Waters
• 06/19/2019

Datacenter solutions vendor Kaloom this week unveiled a new offering the company says will enable the development of "self-driving" datacenter networks.

Kaloom's flowEye is an AI-powered, real-time, in-band network telemetry (INT) and analytics solution designed to trace actual packet routes as they travel through the datacenter.

Kaloom is targeting a growing pain point with this release in an evolving industry. Packets now travel through virtual system infrastructures that include controllers, routers, gateways and security, each of which may be installed on different hardware.

Traditional network telemetry and analytics solutions that rely on sampled or synthetic traffic, or packet-probing protocols, were not built for virtual architectures, and they're not keeping up with demands for real-time visibility into the multiple data types.

The flowEye solution comprises a real-time INT component that collects data via hardware- or software-based sensors and an analytics component (either Kaloom's or a customer-owned third-party solution) that aggregates the collected data, performs data analysis and displays that analysis on a dashboard.

The solution "leverages the capabilities of the P4 programmable ASIC based switches to obtain granular, real-time insight of the network state regarding key metrics pertaining to packets based on actual traffic and their flow as they traverse the network," Kaloom's product Web site explains.

Paul Parker-Johnson, chief analyst at ACG Research, noted in a statement that INT, analytics and automation are well-known for their potential to improve datacenter operations and enhance application deliveries. He sees Kaloom's perspective on analytics and automation in the datacenter environment as "both innovative and forward-looking."

"Kaloom's flowEye innovations significantly increase the granularity of insight that can be applied to pursuing those goals," he said. "And the fact that its functions are programmable into functioning datacenter resources means the outputs from its analytics are available faster than existing solutions, and at lower overall cost, since no additional equipment needs to be installed."

The flowEye solution follows packet routes through the datacenter, tracking where they've been and the impacts of throughput among both virtual and physical elements. The networking nodes along the path use the INT instructions to tell devices what state to collect and write that information into the packet as it transits the network.

One key advantage of this system: The improved granularity provided in real-time facilitates root cause analysis, so network problems can be pinpointed before they occur and corrective actions can be taken.

"Our automated real-time monitoring, analytics, and troubleshooting capabilities will change the way datacenters are currently managed," said Kaloom CTO Suresh Krishnan in a statement. "Kaloom has taken a unique approach and can now provide an industry-first, real-time visibility of the packets and flows for the actual traffic, thus guaranteeing optimum operational visibility for datacenters."

Kaloom is based in the heart of the Quartier de l'innovation in Montreal, Quebec, and at the center of Silicon Valley in Santa Clara, Calif. The company was founded by CEO Laurent Marchand, whose goal was to provide "the industry's most automated and programmable datacenter networking fabric, one that will disrupt how datacenters are built and managed by cloud providers, telcos, and enterprises."

The company is almost certainly going to make a big splash with its flowEye release, but it's probably best-known for its automated Software Defined Fabric (SDF), an industrialized software solution for open networking whiteboxes, which it launched last year. Kaloom targeted its SDF at hyperscale and distributed datacenter environments, where there's a need for low-latency, low-cost, programmable solutions.

Another Kaloom offering, Cloud Edge Fabric, is a fully automated datacenter networking fabric solution specifically designed for simultaneous 4G and 5G applications at the datacenter edge, providing native support for network slicing along with embedded 5G user plane function.