TroubleShooting: Unexpected low TCP speed

Introduction

ByteBlower servers and their assigned switch(es) are by default configured to be able to achieve the maximum supported line rate for that particular interface. However, link speed limitations on your systems under test can introduce reduced throughput when testing with ByteBlower. There are cases when it is necessary to fine tune your ByteBlower System to work optimally with your system under test.

Signs to Fine Tune

When is it necessary to fine tune your system? Consider the following.

Results in your line rate tests max out significantly less than the line rate.
High packet loss / Inexplicable rates.
High Rate of Retransmissions / Inexplicable rates.

Use Case

A PC with a 2.5 GbE NIC is connected to a 1G/2.5G/5G/10G trunk port of a ByteBlower switch. An endpoint test is conducted with Frame Blasting, the test shows 68% packet loss on UDP, and the speed maxes out at 1.7 Gbps, despite being on a 2.5Gbit link. Why is this?

Explanation

The Ports on the ByteBlower switches are capable of auto-negotiation. However, link speed limitations on your systems under test can introduce reduced throughput when testing with ByteBlower. What does this mean in practice?

When executing a traffic test without maximum speed limitations, the ByteBlower server will transmit at the auto-negotiated (maximum line rate) speed (100Gbps, 10Gbps) with the (ByteBlower) switch it is connected to, which is not directly connected to the system under test. That system under test might have negotiated lower link speeds (2.5Gbps, 1Gbps, 5Gbps).This may cause internal buffers of the (ByteBlower) switch to overload and cause high latency and even packet drops.

In a nutshell, although the ByteBlower switch and the device under test negotiate a link speed, the user, when configuring a Byteblower Test Project, must consider the negotiated speed, since as a user you are able to oversubscribe the line.

Also to note: some devices although using 2.5 GbE interfaces may have performance issues due to high buffering, interference, suboptimal configurations, etc.

Solution

Enter the ByteBlower Server Configurator Tool. Using the ByteBlower Configurator Tool, you can be sure not to have oversubscription of the line at any particular ByteBlower Port. This tool allows you to limit the sending speed of the ByteBlower Server to the multiplexing ports (on the ByteBlower Switch). Which allows you to match the max sending/receiving rate for the intended system under test connected at each port of the ByteBlower switch.

Visit this KB to learn how to configure each port: Interface Speed Configuration. It is recommended to work with the Excentis Support team to make these changes.

Keep in mind: you may need to revert/update the changes whenever the DUT connected to that port has been changed.

Advantage of a Flexible System

While other server (such as Iperf) are limited to the negotiated link speed of the receiving port, with ByteBlower, you are able to stress your system under test or DUT, identify bottlenecks, and if needed customise the sending speeds. True flexibility for all your network testing needs.

Why is the behaviour with Iperf different? Iperf traffic comes from a server running on the internet therefore it is subject to buffering along each hop. Further more this connection is aware of the link speed so it will never be able to send faster than the connected/negotiated link speed.