Troubleshooting Low Network Throughput (<10Gbps) on Dell PowerEdge R750

Products

• VMware vSphere ESXi

Issue/Introduction

 

• The Dell PowerEdge R750 nodes in a vSphere cluster exhibit significantly lower-than-expected network throughput.

Bad host 

• Despite having 25Gbps NICs, internal iperf3 tests between VMs on the same host (using the virtual switch backplane) yielded results as low as 8-13Gbps, compared to the expected 30-40Gbps.

Example:

One host achieves ~30.8 Gbps (Expected), while another (the "bad host") caps at ~8.59 Gbps.

Bidirectional tests show uneven results (e.g., 13 Gbps one way, 6 Gbps the other).

•  esxtop reveals high interrupt values on the problematic host as show below:

 

 

 

Environment

Vmware ESXI 

Cause

The root cause is a defective hardware component (likely a faulty CPU or Motherboard/VRM) generating excessive interrupts. This "hardware noise" prevents the CPU from efficiently processing network packets, leading to the massive drop in iperf3 performance even when the traffic is purely internal to the ESXi host.

Resolution

To isolate the physical hardware from the network fabric,  conduct a test using a Uvpsa VM (or standard Linux VM) using the following iperf3 parameters:

iperf3 -c x.x.x.x -w 256k -t 180

• -c: Client mode pointing to the target server IP.

• -w 256k: Specifically defined window size.

• -t 180: Duration of 180 seconds to ensure sustained performance stability.

 

 

Use the esxtop utility to compare a "Good" host against the "Bad" host.

1. Run esxtop from the ESXi CLI.

2. Press 'i' to view interrupt statistics.

3. Monitor the "I" (Interrupts) column while the iperf3 test is running.

 

• Good Host: Low to zero unexpected interrupts during high-load traffic.

• Bad Host: Consistent high interrupt spikes, indicating the CPU is being overwhelmed by hardware-level requests.

 

 

This is hardware issue causing the interrupts, Please Open Support Case with Dell Technologies/H/W vendor to move forward.