Chief Technology Officer, Applications, Barefoot Switch Division
By Changhoon Kim
Getting stuck in traffic is not fun. Not knowing what’s going on or how long you’ll be there is even worse. But it is an apt metaphor for the challenges – and solutions – facing data center networking.
Before the internet, people relied on the radio for traffic alerts. If you heard about a traffic jam and had the option to take an alternate route, the system worked fine. However, if you heard the traffic report after it was too late to take an alternate route, the system failed because the tracking and reporting didn’t happen in real time. Meanwhile, without knowing the exact location of each car and where it was heading, the traffic control system offered limited optimality.
Today, nearly every car has a GPS tracking system that uses the driver’s speed input and location to compile a live view of traffic conditions. The system can then navigate you onto the best route at any moment in time. This is analogous to data center traffic: Without tracking systems to identify network congestion issues in real time, network owners will inevitably experience delays before they have an opportunity to troubleshoot problems.
At Intel, we solve this problem using In-Band Network Telemetry (INT). I discussed the power of INT and real-time telemetry in my recent Micron to Miles briefing. INT is an imperative technology we use to help accelerate network performance, improve network agility and increase network visibility.
Telemetry in Real Time
Unlike alternative telemetry solutions that produce the same lag we see with radio traffic reports, INT collects data on every packet in the network as it happens. This provides a comprehensive understanding of network bottlenecks and steers packets to the least congested paths. Consequently, INT provides unprecedented visibility into packet conditions and offers complete network transparency. In terms of real-time applications such as video conferencing, voice over IP calls and factory automation, INT helps to avoid any loss or delay that could degrade application performance and cause noticeable consequences.
For in-depth analysis of INT information, Intel® Deep Insight™ Network Analytics Software collects, compiles and reports on all telemetry data for a holistic view of the network. This tool processes packet reports in real time, implementing intelligent analytics and data correlation techniques to detect network anomalies. These reports can also be stored for deeper historical analysis to allow data flows to react to other throughput degradation issues. Decisions can be made autonomously at any INT-capable device as well, further reducing the reaction time to congestion issues.
In my Micron to Miles presentation, we demonstrated an exciting new mode called “hos-based INT.” This capability is designed to report on network conditions from a viewpoint outside the network. For example, internet game companies depend on reliable internet speeds for a positive customer experience. With host-enabled INT, they can gauge performance from one end of the network to the other and determine whether an issue is coming from the network or their software application.
INT and Intel Deep Insight Network Analytics Software provide a huge leap in reducing network congestion and anticipate the fastest path for data packets to reach their destination. These technologies can empower our customers to gain an in-depth understanding of challenges facing their networks, which is, and will continue to be, ever so critical as the demand for data continues to grow.
Changhoon Kim is an Intel Fellow and chief technology officer of Applications in the Barefoot Switch Division of Intel Corporation.