When 10 GB isn’t actually 10 GB

Matrium Technologies Pty Ltd
Thursday, 20 September, 2012


Over the years Matrium Technologies has been asked to benchmark many customer networks using our suite of traffic generation tools, capable of producing 1 GB through to 100 GB of network traffic. Having performed thousands of test cases for hundreds of customers, we know that results vary. Every customer’s network is different and therefore results are never the same. One thing is the same though - the infrastructure you’re about to purchase won’t perform as per the data sheet.

To show you what we mean, have a look at the results from testing performed for one of our recent engagements. In this example our customer had paid vendor ‘X’ for an end-to-end 10 GB-capable infrastructure in its brand new data centre.

Our goal was to test a number of paths/flows through the data centre with live traffic, in essence stress testing the design. All of these tests were to be against 10 GB infrastructure, which has become the norm these days. The expectation was to achieve at least 6.5 GBps throughput, an acceptable target considering expenditure and future capacity considerations built into the design.

Prior to our engagement the customer had ensured with the vendor, who also performed integration on the project, that the configuration was as close to operational as possible. The customer also thought it prudent to test the design as part of their assurance and handover procedure.

Our suite of testing procedures highlighted packet drops occurring at very low throughput - far less than what the 10 GB link was supposed to be able to support. Packet loss wasn’t the only issue found with this testing; latency measurements were way too high for traffic traversing the same data centre, which would cause severe impact to latency sensitive applications on that network.

Frame size Max TxPut Gbps Percent loss Average latency (µs) Average jitter (µs)
64 0.1 0.39 435.67 84.38
128 0.2 0 414.85 75.51
256 0.4 0 467.72 75.03
512 0.5 0 446.86 59.8
1024 1 0 550.85 81.73
1260 1.25 0 956.15 155.02
1518 1.25 0 1143.31 184.18

This network had a fundamental network issue and it was pinpointed while testing. Matrium worked with the vendor to reproduce the issue in its labs, resulting in patches being written and the customer’s data centre bottleneck being fixed; all at zero impact to the production environment, while producing a result that saved network issues once it had become operational.

If you have a network whose performance is critical or latency sensitive applications, contact Matrium’s Professional Services team to verify your network.

By Trent Owens, Business Development Manager, Matrium Technologies

Related Articles

Powering data centres in the age of AI

As data centres are increasingly relied upon to support power-hungry AI services and...

Smart cities, built from scratch

With their reliance on interconnected systems and sustainable technologies, smart cities present...

Smart homes, cities and industry: Wi-Fi HaLow moves into the real world

Wi-Fi HaLow's reported advantages include extended ranges and battery life, minimised...


  • All content Copyright © 2024 Westwick-Farrow Pty Ltd