Responding to High TCP Retransmission Rates in Bitfusion

Products

• VMware vSphere ESXi

Issue/Introduction

Symptoms:

Bitfusion depends on fast and reliable networking. Ordinarily, the TCP retransmission rate should be very low and near zero. If TCP retransmissions are significant, it can result in poor performance. If there is poor performance in the Bitfusion environment, check for high TCP retransmission.

Environment

VMware vSphere Bitfusion 2.x

Resolution

To check for high TCP retransmission rates, use the iperf3 utility. This utility may need to be installed on clients and servers if not already present. iperf3 will send traffic between endpoints and report on networking statistics. To use iperf3 to check for network statistics run the following commands:

Bitfusion Server	Bitfusion Client
iperf3 -s	iperf3 -c <server IP address>

All results are examined from the client side of the test. After allowing the utility to collect data and show results, you can terminate the server by running “CTRL+C” in the terminal where the server is running.

If TCP retransmission activity is frequently or constantly observed, this is the likely cause of poor Bitfusion performance. To help address TCP retransmission issues, here are potential avenues for further investigation:

• Ensure that all physical-layer components (cables, optics, etc.) are in good condition and error-free.
  • Errors can be observed from network switch ports or physical NIC statistics
•  Ensure that applicable NIC drivers and firmware are up-to-date and compliant per the VMware Compatibility Guide.
• Ensure that switch buffers are not exhausted by examining the appropriate switch ports or line cards on the physical switches
• Ensure that ring buffers on the physical network interfaces on the ESXi host are not exhausted.
  • For more information, see Network receive traffic faults or poor performance in ESXi 
• Ensure that ring buffers on the virtual network interfaces in the Bitfusion VMs are not exhausted.​​​​​​​
  • It is recommended per Best Practices for VMware Bitfusion 2.0 that LRO/TSO and coalescence be disabled on virtual interfaces for Bitufsion machines.