(External) Bridging Performance (Core & Scale)

[mixed]

Hi everyone,

Just wondering what the current state of (external) bridging is -- in terms of expectable performance -- on TrueNAS Core and Scale. In reading various forum discussions and concerns, was wondering what the current perspective would be on the following 2 questions:

1. Has anyone had a chance to compare performance on 10Gbps Intel NICs (Base-T or SPF+) in bridge mode? How significant a drop in performance is generally to be expected (on, let's say, something like bridging 3 ports on a quad port card [for hooking up to XCP-ng Hosts for access to NFS Shares])?

2. Is turning off NIC offloading presently recommended on Scale like it is on Core (some Debian How-To's also suggest this for performance)?

Thanks.

 

[mixed]

If such an (external) bridging setup is generally not attempted, is bridging just something to generally avoid on Core and/or Scale?

 

[mixed]

[Solved]

In case anyone is similarly curious in the future:

Testing on two intel hosts of differing hardware with intel and Broadcom 10Gbps Base-T NICs (on TrueNAS Core 13.0-U6.1) shows no appreciable slowdown in bridge mode:

XCP-ng Host 1:

XCP-ng Host 2:

Results are with each host running 1 VM at the same time (connecting to NFS Storage over same Storage network).

(MTU 9000)

These results were very pleasantly surprising.

 

[Patrick M. Hausen]

Testing on two intel hosts of differing hardware with intel and Broadcom 10Gbps Base-T NICs (on TrueNAS Core 13.0-U6.1) shows no appreciable slowdown in bridge mode

Expected

The FreeBSD Foundation sponsored a complete rewrite of the bridging code in FreeBSD:

5x if_bridge Performance Improvement – FreeBSD Foundation

freebsdfoundation.org

In private conversation Kristof Provost told me that in any scenario the bridge interface will probably not be the limiting factor, anymore.

 

[mixed]

Expected

The FreeBSD Foundation sponsored a complete rewrite of the bridging code in FreeBSD:

5x if_bridge Performance Improvement – FreeBSD Foundation

freebsdfoundation.org

In private conversation Kristof Provost told me that in any scenario the bridge interface will probably not be the limiting factor, anymore.

Danke Patrick for confirming that this result can be widely expected (and the informative link).

Kristof Provost's work on these matters is much appreciated.