Abstract
Network performance is key to the correct operation of any modern data centre infrastructure or data acquisition (DAQ) system. Hence, it is crucial to ensure the devices employed in the network are carefully selected to meet the required needs. Specialized commercial testers implement standardized tests [1, 2], which benchmark the performance of network devices under reproducible, yet artificial conditions. Netbench is a network-testing framework, relying on commodity servers and NICs, that enables the evaluation of network devices performance for handling traffic-patterns that closely resemble real-life usage, at a reasonably affordable price. We will present the architecture of the Netbench framework, its capabilities and how they complement the use of specialized commercial testers (e.g. competing TCP flows that create temporary congestion provide a good benchmark of buffering capabilities in real-life scenarios). Last but not least, we will describe how CERN used Netbench for performing large scale tests with partial-mesh and full-mesh TCP flows [3], an essential validation point during its most recent high-end routers call for tender.
Highlights
When evaluating the performance of network devices we need to ensure they have the required features in terms of management and protocols and, more importantly, in terms of raw performance.Given the diverse nature of CERN’s networks, we need to ensure that devices can correctly cope with various traffic patterns
A full-mesh traffic distribution [3] is most likely the best benchmarking pattern for a network device to be employed in a multi-purpose data-centre;
We will introduce Netbench and describe how it addresses the shortcomings of specialized commercial testers
Summary
When evaluating the performance of network devices we need to ensure they have the required features in terms of management and protocols (software features, with a fast enhancement cycle) and, more importantly, in terms of raw performance (hardware features, with a much slower enhancement cycle). (e.g. buffer depth) are not exercised by these tests, and performing a full test for a high end device with many ports requires a large number of expensive commercial testers. We will introduce Netbench (the network testing framework that we have developed, which uses commodity servers and the iperf3 [4] traffic generator) and describe how it addresses the shortcomings of specialized commercial testers (i.e. the cost of large-scale setups, and the ability to inject realistic traffic patterns that exercise buffering capabilities). We will present how CERN has successfully benchmarked high-end routers using an affordable combination of a small scale setup with specialized commercial testers (two ports) and a Netbench setup with 64 servers for large scale traffic injection
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call