User experience - turris 2.5GBASE-T sfp

setup
ISP <10G> CRS305 <2.5G> MOX

Now, I connected all together and ran some speed tests.
But as you can see above I need to run the tests with the MOX as client to public speedtest services to see anything above 1G, as my LAN side only have a 1G port.
This is far from optimal but for this case it will have to be acceptable.

And as my baseline tests with 1G sfp optics have been stable enough I think it should be ok to make rough estimates.

#Baseline 1G sfp

turris:~# speedtest-netperf.sh -H netperf-eu.bufferbloat.net -t 30
..............................
 Download: 915.01 Mbps
..............................
 Upload: 922.13 Mbps

#2.5G sfp

turris:~# speedtest-netperf.sh -H netperf-eu.bufferbloat.net -t 30
...............................
 Download: 487.49 Mbps
.............................
 Upload: 2322.99 Mbps

Thats 2.3Gbps upload, how cool is that on a litle MOX, looks like it maxes out at about 1600 whith SQM (simple) enabled (100%CPU)
however the download is not really what I expected, and other tests confirmed that it will not get anywhere close to the 1G baseline.
So I check logs and counters and there is some disturbing stuff:

turris:~# ethtool -S eth1
rx_discard: 27382 
rx_overrun: 400
turris:~# ifconfig eth1
RX packets:9240038 errors:5163

kernel logs have a lot of below with different sizes
“mvneta d0040000.ethernet eth1: bad rx status 0f830000 (overrun error), size=704”

When running iperf3 locally I get stable 1G performance with the same sfp:s with out any rx_overrun or rx_discard:s

server <1G> CRS305 <2.5G> MOX

First I was sure my mikrotik is to blame but I’m not so sure anymore.

Any ideas or similar experience?

1 Like

How did you come across that module? Was it announced somewhere?
I’d really like to see an iperf measurement of a connection between two MOX or two TO…

Yes, I should have menrioned that in the post above.
No iperf between moxex though, that would be interresting

Announcement

Iperf

1 Like

Looks like there have been a bit of work on MOX sfp side in 5.3.4.
I will give it a try again in a couple of days

Nope, same experience with 5.3.4.

I guess next step would be to run iperf with direct connection between a mox and some other 2.5G/10G capable device, as @ssdnvv suggests.

Sadly I do not have such device at the moment.
We’ll see what the future holds.

But if anyone have the gear to run such test please share the detailes.

From practical perspective, I don’t find it common for a router to consume much bandwidth for itself, but rather route almost all of it towards LAN(s)…

1 Like

@vcunat Yes I fully agree, if thoughput tests is your goal.

Client <2.5G> MOX <2.5G> Client
Would be a better test for that porpose

Or rather
Client <2.5G> MOX <3x1G> 3xclients
As we only have one 2.5G capable port on the MOX.

Well that will be even more problematic for me as I only have A+D modules on my mox.

Have a bit of a struggle to justify buying stuff just for test/lab purpose, but lets see what the future holds