Throughput Insights from Certification of a Large Network

 

Stephen Kingham - Integ

 

Abstract

The customer complains that the Service is slow, The Network engineer tells the user that the network is under utilised and thus it must be the Server or the Workstation”.  How often have I said the same thing only now after certification of a large data network it turned out that it is the network that is causing slow response!  All the MRTG graphs and SNMP could not detect that the network was causing the problem.  Another similar phrase is from a researcher wondering why the 1/10Gbps network is slow!

This presentation, a case study, gives insights from certification of a large network that could not sustain throughput of more than 50% of the bandwidth ;-(.  The introduction of QoS configurations has complicated the configuration of switches and routers and the queue depths needed to sustain the throughput is what caused the problems.  The vendor defaults were nothing more than crazy.  This presentation shows the tools used to find the problem, why MRTG and SNMP could not see the problem, and the calculations to set up the queues in the QoS configuration.

Bio

Stephen Kingham from 1998 to 2006 architected and project managed the VoIP and VIDEOoverIP network used by CSIRO and the Australian Universities. During this time he lead the introduction of VoIP, IP Telephony, VIDEO over IP, and SIP based services. Up to mid 2007 Stephen was a lead Design Engineer on Centrelink's new data network and specialised on the Quality of Service. 

To the present he has been working with The Australian Department of Foreign Affairs on the support and deployment of a world wide IP Telephony network.

Stephen has been in the Telecommunications industry since 1980 and has several tertiary qualifications.  He is also active in National and International forums and was the founding chair and presently the Secretary of the Asia Pacific Advanced Networks (APAN) Working Group on SIP and H.323.