Current speeds on Quic 2Gbps Hyperfibre

Forums › Vetta and Quic › Current speeds on Quic 2Gbps Hyperfibre

1 | 2

godber

75 posts

Master Geek
+1 received by user: 46

ID Verified

Trusted

Lifetime subscriber

#3157727 9-Nov-2023 17:29

Also on Quic I have the same issue about 2gbps down, depending in the Speedtest server, Chorus Fibre Lab being the most reliable. While upload is around 450mbps.

Router is a Mikrotik CCR2116-12G-4S+ with a Mikrotik RJ45 SFP+ 10/100/1000M 2.5G/5G/10G copper module installed It has 16 cores and 16 GiB of memory so resources are not an issue. However PPPOE only uses one core, that hits about 83% on download when at 2gbps but on upload never goes above 15%.

Server is a dual Xeon HP DL380 G9 with 192 GiB memory. It is running ubuntu Linux on the hardware not as a VM and am using the Speedtest CLI from https://www.speedtest.net/apps/cli.

So hardware at my end should not be an issue even with PPPOE, I once hit just over 1gbps upload so likely all setting at the ISP/Chorus end are correct.

Will have to wait for PoE (DHCP) to see if that solves. But even on good hardware PPPOE seems to be an issue.

Godfrey
Auckland/Coroglen, New Zealand
Quic Broadband - 4G Hyperfibre

Referral Link:
Quic (use R71004E9PVBJ on checkout for free setup)

RunningMan

9185 posts

Uber Geek
+1 received by user: 4838

#3157730 9-Nov-2023 17:42

@godber 1500 MTU on the PPP connection?

godber

75 posts

Master Geek
+1 received by user: 46

ID Verified

Trusted

Lifetime subscriber

#3157731 9-Nov-2023 17:49

RunningMan:

@godber 1500 MTU on the PPP connection?

@RunningMan 1492 MTU on the PPP connection. It autodetects.

Godfrey
Auckland/Coroglen, New Zealand
Quic Broadband - 4G Hyperfibre

Referral Link:
Quic (use R71004E9PVBJ on checkout for free setup)

RunningMan

9185 posts

Uber Geek
+1 received by user: 4838

#3157736 9-Nov-2023 19:11

@godber set max MTU & MRU to 1500 for the PPPoE connection. Assuming untagged then set the MTU of the physical ethernet interface to 1508. Quic supports full 1500 on PPPoE.

BMarquis

465 posts

Ultimate Geek
+1 received by user: 577

Trusted

Chorus

Lifetime subscriber

#3157738 9-Nov-2023 19:50

That 8Bytes of inefficiency by not using 1500B PPP MTU only translates to 0.53%.
So on a 2gbps service, you only loose 10mbps.

While it’s good to use optimal settings…
I don’t think MTU is contributing in any meaningful way to the issues that users are reporting here.

godber

75 posts

Master Geek
+1 received by user: 46

ID Verified

Trusted

Lifetime subscriber

#3157823 9-Nov-2023 23:25

RunningMan:

@godber set max MTU & MRU to 1500 for the PPPoE connection. Assuming untagged then set the MTU of the physical ethernet interface to 1508. Quic supports full 1500 on PPPoE.

@RunningMan Many thanks that made a substantial difference. Download is still at 2Gbps but upload is now about 1.6Gbps

Interesting on upload PPPoE seems to use multiple cores as about 6 cores hit about 35% While download with the new MTU & MTR hit 93%.

The change happened automatically to PPPoE when I changed the physical ethernet interface to 1508. But I have made the change to max MTU & MRU to 1500 for the PPPoE connection as well.

Godfrey
Auckland/Coroglen, New Zealand
Quic Broadband - 4G Hyperfibre

Referral Link:
Quic (use R71004E9PVBJ on checkout for free setup)

New World

Shop on-line at New World now for your groceries (affiliate link).

kennedybaird

63 posts

Master Geek
+1 received by user: 50

ID Verified

#3157836 10-Nov-2023 06:52

RunningMan:

Easy to check - look at the CPU usage when uploading and see if 1 core is saturated.

They aren't getting anywhere near saturated. Max 25%, shared across the 4 cores.

This is what I updated my MTU values to a while back (you gave me this advice when I was trying ZeroNet):

Is that correct, or should sfp be 1508? SFP port is incoming from ONT

nzkc

1634 posts

Uber Geek
+1 received by user: 1041

#3157939 10-Nov-2023 10:12

I think it should be 1508. I'd definitely try it! Easy to switch back if it makes zero difference.

RunningMan

9185 posts

Uber Geek
+1 received by user: 4838

#3158073 10-Nov-2023 15:13

@kennedybaird that's fine, assuming you're running a tagged connection. In fact it'll work OK on untagged as well. You wouldn't get the full 1500 byte PPPoE frame size otherwise. In System/Resources are you looking at the individual CPU core loads or the combined because 25% could be one core saturated.

RunningMan

9185 posts

Uber Geek
+1 received by user: 4838

#3158074 10-Nov-2023 15:16

BMarquis: That 8Bytes of inefficiency by not using 1500B PPP MTU only translates to 0.53%.
So on a 2gbps service, you only loose 10mbps.

Totally agree with that, assuming you can saturate the link, but that doesn't seem to be happening in this case. It was more if the router was hitting a CPU limit or something due to having to fragment packets.

BMarquis

465 posts

Ultimate Geek
+1 received by user: 577

Trusted

Chorus

Lifetime subscriber

#3158077 10-Nov-2023 15:27

Fair call, I hadn't considered that.
Or worse, just simply not fragmenting and ditching frames over 1492 :D

Apple TV

Stream your favourite shows now on Apple TV (affiliate link).

kennedybaird

63 posts

Master Geek
+1 received by user: 50

ID Verified

#3158095 10-Nov-2023 16:32

RunningMan:

@kennedybaird that's fine, assuming you're running a tagged connection. In fact it'll work OK on untagged as well. You wouldn't get the full 1500 byte PPPoE frame size otherwise. In System/Resources are you looking at the individual CPU core loads or the combined because 25% could be one core saturated.

Tagged - I set VLAN ID 10 when setting up. This is what you mean?

As for the CPU - it's max approx 20% across all four cores when downloading speedtest, max a single core went to was 31%.

For uploading, it's doing the same but the figures are lower all round. No single core is saturated.

RunningMan

9185 posts

Uber Geek
+1 received by user: 4838

#3158100 10-Nov-2023 16:53

Yes re the tagging - you obviously requested a tagged connection with Quic as default is untagged.

That's interesting when compared to what @godber is seeing on different Mikrotik hardware. @godber is close to saturating a single core on the CCR2116 with 2GHz ARM processor whereas the RB5009 (1400MHz) is nowhere near that. Haven't checked if the RB5009 switch chip will do the VLAN tagging in hardware, but that could explain the extra load across other cores if it doesn't (or not using fasttrack or more complex firewall rules too). I wonder if the CCR2116 reduces the clock speed unless multiple cores are getting loaded up?

Be interesting to look at tools/profile to see what's eating the CPU time.

1 | 2