INTRODUCTIONCloud computing is significantly transforming the way traditional IT and telecom infrastructure is built and deployed. OpenStack (http://www.open-stack.org) is a free and open-source cloud com-puting platform, backed by several industry-leading players, which offers tools and technolo-gies that work with both virtual as well as bare-metal resources.Visibility into Top-of-the-Rack (ToR) switches is the critical link in managing cloud networks. Although cloud networks offer a lot of flexibility, overlay technologies reduce visibility into traffic flows and make it harder to troubleshoot.Traditionally, network congestion has been asso-ciated with switch ports being utilized at close to line rate. In a congestion scenario, packets can be dropped by the switch or flows may be back pres-sured due to the lack of buffer space. However, more recent analysis has uncovered the exis-tence of microbursts, where rapid bursts of pack-ets at line rate can temporarily overflow the switch buffers and cause packet loss or backpressure. Typically, these microbursts do not last long enough to be detected by traditional switch count-ers such as SNMP or port statistics. In cloud envi-ronments, increased complexity and reduced visibility make detecting these microbursts all the more difficult.Ceilometer is a relatively new component of OpenStack. The Ceilometer project originally aimed to deliver a unique point of contact for bill-ing systems to acquire all of the measurements they need to establish customer billing, across all OpenStack core components. It has evolved to be the means by which all instrumentation and moni-toring-related data can be collected and stored in an OpenStack environment.While OpenStack Ceilometer reports course-grain data about tenant (overlay) networks, it does not have the ability to inspect data about the underlay network provided by physical switching hardware in the OpenStack cluster.

BroadView Analytics and OpenStackOpenStack clusters that deploy physical switches based on Broadcom's BroadView Instrumentation software enable increased visibility in the opera-tion of the underlay network in a way that is com-patible with OpenStack. The collector can either pull instrumentation data from the agent or have the data pushed by the agent in an asynchronous manner. In the solution described here, a BroadView agent running on the switch sends statistics reports asynchronously to a collector application running on the Open-Stack controller. The reported data includes exhaustive counters covering the buffer usage of device, interface, and shared service pools. In our demo, a standard traffic generator is used to play traffic into the switch at a high rate, to simulate the microburst scenario discussed above. When buf-fer utilization becomes sufficiently high, the agent reports this state in the form of absolute buffer and percentage values for the device and shared service pools as configured by the collector. Decreasing the amount of data flowing from Ixia lowers the buffer utilization, which is reflected in smaller values reported by the BroadView agent back to the collector for the statistics being moni-tored.

CeilometerNotification

Agent

CeilometerCollector

OpenStack Notification Bus (OSLO)

BroadView™Plugin

BroadView

OpenStack

BroadView Collector - Common Java Module

BroadView Agent

Trident II Platform

TrafficGenerator

The benefits of using BroadView include:• The ability to visualize advanced

telemetry to monitor and detect microburst-like congestion sce-narios.

• Detailed statistics displayed in an easy-to-use graphical format.

• User control of which statistics are reported.

• Easy integration of the Instrumen-tation agent with network OS.

Having a proactive statistics tool that can give details per queue can be a great help in monitoring and maintaining network stability and encouraging optimum network usage, particularly in those designed with OpenStack.

TECHNICAL BRIEF

BroadViewIncreased Efficiency and Simplicity to OpenStack Cloud with BroadView™ Network Agent

Benefits of BroadView™ in the OpenStack Ceilometer Project

© 2016 Broadcom. All rights reserved. Broadcom®, the pulse logo, Connecting everything®, the Connecting everything logo, and Avago Technologies are among the trademarks of Broadcom and/or its affiliates in the United States, certain other countries and/or the EU. Any other trademarks or trade names mentioned are the property of their respective owners. Broadcom reserves the right to make changes without further notice to any products or data herein to improve reliability, function, or design.

ABOUT BROADCOMBroadcom (NASDAQ: AVGO) is a diversified global semiconductor leader built on 50 years of innovation, collaboration and engineering excellence. Broadcom’s extensive product portfolio serves multiple applications within four primary end markets: wired infrastructure, wireless communications, enterprise storage and industrial & others. Broadcom is changing the world by Connecting everything®. For more information, go to www.broadcom.com.BroadView_OpenStack-TB101-R • March 22, 2016

The collector application is Broadcom-developed software that uses industry standards, such as JSON-RPC, and a REST-like API to configure the switch-res-ident agent to receive reports of buffer-related statistics from a compatible switch. The collector is written using a Broadcom-developed portable Java library that pro-vides support for connecting to and man-aging the BroadView agent on the switch. The collector application publishes data reported by the Java library to Open-Stack's OSLO messaging bus. A Broad-View plugin to the Ceilometer Notification agent monitors the OSLO bus for this data and converts it into data that is under-standable by Ceilometer, republishing the results back onto the OSLO bus. Open-Stack Ceilometer then reads this data from the bus and ensures that it is placed in the Ceilometer database, where it can then be read by visualization or analytics applications.Ultimately, the data written to the Ceilome-ter database contains reports on buffer uti-lization, which can be used to detect microbursts associated with congestion in the network. The data is expressed as a percentage of buffer utilization, and includes timestamp information that indi-cates when the data was collected.Ganglia, a widely known monitoring sys-tem, is used in this solution to provide visualization of the BST statistics that have been written to the Ceilometer data-base. A small helper agent reads data from the Ceilometer database and repub-lishes it to Ganglia.

Additionally, the OpenStack dashboard, Horizon, has been modified to display the BroadView BST counter data that is pres-ent in the Ceilometer database. This data is viewable along with other Ceilometer data collected by the OpenStack cluster in the form of an X-Y line graphic, where X represents time and Y represents percent-age buffer utilization.

DEMO COMPONENTS• A server (x86-64) running Ubuntu Desk-

top 14.04 or later• OpenStack stable/juno configured with

Ceilometer support, installed using devstack

• Ganglia (the default version for Ubuntu)• A Broadcom BCM56850-based switch

node running the BroadView BST appli-cation over the SDK

• A traffic generator/sink such as Ixia or Spirent (three 10G/40G fiber ports)

• A collector application based on BroadView's Java-based API for accessing BST data and publishing it to the OpenStack OSLO bus.

• A Ceilometer agent plugin that converts data written by the collector to samples understandable by the Ceilometer.

• A helper application that reads BST data from the Ceilometer and publishes it to Ganglia.

DEMO FLOW• The BroadView agent is started on the

switch.• OpenStack is launched (Single node,

nova, neutron, Ceilometer and other components) using devstack

• The BroadView Java Collector applica-tion is started in a console window. It configures BroadView BST.

• A helper app is run in a console window that consumes BST data from the Ceilo-meter and publishes it to Ganglia.

• A standard traffic generator is used to play microbursts into the switch.

• The Ganglia web UI is used to inspect the Ceilometer data related to the BroadView switch (as well as the host OpenStack is executing on), giving a complete picture of the OpenStack infra-structure and underlay.

• The Horizon dashboard application is used to list the BST meters as well as to show the Ceilometer counters.

FOR MORE INFORMATIONBroadView Product Page:www.broadcom.com/products/ethernet-communication-and-switching/switching/broadview