Cavium - Semiconductor Solutions and Server Innovation

Cavium / XPliant® Packet Trakker™ Programmable Telemetry Solution

Overview

The need to manage the interaction between data center network infrastructure resources and constantly changing application workloads has never been more important to maintaining peak data center performance.  To do this effectively a comprehensive suite of network telemetry tools that monitors network performance and status in real-time and at scale is required.  The network telemetry tools must be able to provide real-time high resolution network diagnostic information to data center network analytics engines and operators to enable the dynamic remediation operations that are used to continuously improve network efficiency and application quality of experience.  Absent this real-time high resolution network diagnostic information the task of capturing an event that disrupts application performance and identifying its cause can be very challenging and will often lead to inaccurate or sub-optimal solutions to data center application performance issues.  High performance data center network management requires high resolution real-time visibility into network node resource utilization as well as end to end network performance metrics to ensure that data center network infrastructure resources are operating at peak efficiency.

Legacy Solutions

Legacy network telemetry solutions based on diagnostic probing and switch counter data polling have become insufficient to meet the demands of cloud scale data center networks.

Legacy network telemetry solutions have the following limitations:

  • NOT event-triggered – important events can be missed, if they happen to occur between the sample or are very transient in nature.
  • NOT real-time – polling is typically done every few seconds which is eternity in today’s high performance networks that utilize 100GE ports capable of transmitting ~150M packets in one second.
  • NOT granular – complete telemetry data is NOT available with per flow resolution making it very hard to measure specific application processing metrics. Without accurate accounting for these metrics switching behavior cannot be properly analyzed and the network cannot be properly configured to deliver optimal application performance.
  • Limited hardware assistance – without the proper telemetry primitives integrated into the switch, software is required to perform telemetry data extraction. Software based telemetry data extraction often leads to inadequate data resolution and scale.

Packet Trakker

Packet Trakker
Packet Trakker is a comprehensive data center infrastructure performance management tool suite that combines the flexibility provided by the XPliant Packet Architecture (XPA™) with innovative integrated telemetry primitives and an intuitive display application to provide high resolution, scalable network monitoring. Packet Trakker enables data center networks to be monitored more closely and accurately than ever before utilizing real-time programmable event driven data capture and display capabilities. Networks built using XPliant switches can use Packet Trakker to provide end to end data center infrastructure performance management.

  • Maintenance of a detailed history of switch resource utilization data and network performance logging that is used to perform performance validation
  • Exportation of detailed real time switch utilization and network performance data that can be used to identify network resource contention patterns and isolate which workloads are causing resource contention
  • Enabling the use of programmable instrumentation data gathering contexts that enable fine grained control of how telemetry information is organized for consumption by visualization tools and analytics engines
  • Monitoring of system health and immediately alarms system management applications to circumstances that can cause data center performance degradation and equipment failure
  • Microburst detection and identification of flows that destabilize application performance
  • Latency fluctuations that negatively impact application behavior predictability
  • Packets loss that is destructive to applications performance, which can be avoided through notification of pre-congestion in transmit queues
  • Network utilization and increased network latency, though providing path distribution and latency metrics per flow
  • Inconsistent configurations or packet processing exceptions

Packet Trakker Benefits

Packet Trakker integrates with Network Operating Systems (NOS) and analytics engines, to deliver a comprehensive set of telemetry features that monitor network status in real-time and at scale. This technology assists data center network operators in dynamic remediation operations that continuously improve network efficiency and application quality of experience. Furthermore, the XPliant programmable switch architecture is future proofed for the continuously evolving world of Software-Defined Networking.



Presentation


View PPT presentation

Featured Video

View All