L2TP performance

L2TP related questions
Post Reply
tj-19
Posts: 36
Joined: 10 Jun 2019, 20:42

L2TP performance

Post by tj-19 »

Hi all,

I was just wondering if anyone is using ACCEL with L2TP and 1 Gbps connections.

I have a test server running using a 40 Gbps - XL710-QDA1 NIC.

The CPU usage is very low and I am using 8 cores.

But the throughput is only around 300 Mbps.

I've tried setting NIC optimisation, such as ethtool -K enp1s0 gso off gro off tso off

Is there any support for DDP (Dynamic Device Personalization) ?

Any ideas?
dimka88
Posts: 866
Joined: 13 Oct 2014, 05:51
Contact:

Re: L2TP performance

Post by dimka88 »

Hi @tj-19, I think in this case it will related to CPU frequency. Try to disable HT, set CPU policy (scaling governor) to performance. Also try to disable mitigations, like meltdown and spectre 1 and 2
tj-19
Posts: 36
Joined: 10 Jun 2019, 20:42

Re: L2TP performance

Post by tj-19 »

Hi,

The CPU is a Xeon E5-2440 which runs at 2.40 GHz.

I disabled Hyperthreading, set the SG and disabled the suggested mitigations.

In all cases the overall CPU is very low.

What is the relation between the frequency and performance in this case?

Many thanks for your help.
dimka88
Posts: 866
Joined: 13 Oct 2014, 05:51
Contact:

Re: L2TP performance

Post by dimka88 »

Hi @tj-19, CPU frequency related to flow speed. Did you try to test speed with speedtest multiflow mode?
Could you show `top` command output and press `1` when you doing test?
tj-19
Posts: 36
Joined: 10 Jun 2019, 20:42

Re: L2TP performance

Post by tj-19 »

Hi,

Sorry for the delay in my reply.

Please see TOP output as requested, in this case I am passing CPU's 0-7 to ACCEL.

The CPU usage in all cases seems very low.

I’ve been testing ACCEL using a wholesale 1 Gbps FTTP L2TP connection which terminated directly to ACCEL-PPP.

I test using speedtest.net and compare the result with a test server connected to the same backend transit.

I get very different throughputs. To some servers I get full speed, to others I get just under 100 Mbps when using ACCEL.

If I test with my test server connected to the same transit, I get much faster speeds in nearly all cases.

I’ve tried tweaking TCP with no improvement and performing all the usual NIC tweaks.

At first, I thought it was an MTU/ MRU issue and made sure these were disabled in the PPP section.

Any Ideas?
Attachments
top.jpg
top.jpg (194.51 KiB) Viewed 17119 times
Post Reply