Strange problem running under ESX

[poodad]

When I run OPNsense (latest version) under ESX 6, it works ok for a few minutes, but after 10 minutes or so performance goes south. I see speedtest.net and fast.com results drop, and the latency reported by speedtest.net (using the same speedtest destination server) doubles or triples.

If I go into the shaper and just create a pipe, performance drops even more.

OPNSense shows very little CPU used.

The same setup runs Sophos SG UTM, Sophos XG, and pfsense without issues.

I use E1000 virtual NICs. I've insured that all the NIC offloading is turned off in OPNSense (although I've tried turning it on with no difference).

My ESX server is a quad core AMD with 16GB ram and two NICs. One NIC connects to a Cisco 800 series DSL router, the other NIC connects to my internal network (one subnet, no VLANs). Each physical NIC connects to a an inside and outside vswitch, and the OPNSense virtual has two vnics - once connected to the inside vswitch, and one connected to the outside vswitch.

My DSL is 3Mb/768k. With either of the Sophos firewalls or pfsense, speedtest.net and fast.com will show very close to those numbers.

OPNSense will start off at close to those numbers, but within 10 minutes will drop to 1Mb. If I create a traffic shaper pipe (no queues, no rules), the speed drops to 100kb. If I delete the pipe, the speed stays 100k until I reboot.

Any ideas?

[bartjsmit]

Create a new OPNsense VM with the cleanest possible configuration - i.e. simple outbound NAT and the os-vmware plugin. See if this VM has the same problem.

Bart...

[poodad]

Did that - same issue.