p-state question for Skylake 6140

[NSugi4]

Our 6140 listens Silicom bypass card. which uses Intel chipset (XL550) which drives total four n/w monitor port.

I have a lab setup to run ‘inline bridge’ traffic with Avalanche traffic generator.

+----------+ +-----------+ +----------+ | Avalanche| | 6140 | | Avalanche| | client |-----=eth6 eth7=------| reflector| +----------+ +-----------+ +----------+

6140: CentOS7 kernel 3.10.0-957.el7.x86_64 (and its updates)

Silicom bypass Eth6/eth7 : OOT i40e Intel driver 2.7.79 (downloaded from intel.com)

1) Default, Intel p-state active mode with H/W assisted

=============================================

• default installation, Intel P-state (intel_pstate driver) is working as h/w assisted active mode.
• Using ‘performance’ CPU frequency governor to reach maximum through put.

Performance data shows: inline traffic throughput (TPS) is ~ 1.9 Gbps, traffic (HTTP) response time is ~600 msec

Notice several spikes and jitter like through put distribution. 

2) Intel p-state passive mode

========================

Based on the linux (kernel.org) documentation, I have set passive mode to intel_pstate driver, using ‘performance’ governor.

• Boot Linux kernel with intel_pstate=no_hwp parameter
• Set ‘passive’ to /sys/devices/system/cpu/intel_pstate/status
• Set ‘performance’ to /sys/devices/system/cpu/cpu[0-9]*/cpufreq/scaling_governor

Performance data shows little high TPS ~1.95 Gbps, but traffic response time is ~750 msec.

== == ==

Here is my questions to get answer or suggestion from Intel

• Using default (h/w assisted intel_pstate driver, performance governor), our data does not show the best TPS. Is there any reason behind it?
• Using passive mode of intel_pstate, our data shows larger response time. Is there any reason behind it?
• Is there any microcode update, or firmware update, to improve performance or the ‘jitter’ like behavior ?
• Any recommend or best practice about the Intel p-state technology on 6140 ? 

We are using ‘default, active mode with h/w assist’ for now.

Any helps are welcome. Let me know 

Thank you,

-Naoya Sugioka

[Emeth_O_Intel]

Hi Naoya,

Thank you so much for contacting us about this.

I was reviewing the information provided and I would like to ask you some details in order to have a better undestanding of this scenario. So, could you please so kind and let me know more details of your system?

Such as:

A) Motherboard and BIOS Version.

B) Processor and the Microcode Data File version. Please run the following command:

# dmesg | grep microcode, then provide us the output in order to verify the version.

C) What is the purpose of this server?

D) Are you trying to mesure the transactions per second (TPS) of the network or an specific application?

E) Could you please so kind and provide us more information about this implementation that you are trying to accomplish?

Regards,

Emeth O.

Intel Customer Support Technician

Under Contract to Intel Corporation

[Emeth_O_Intel]

Hi Naoya, I am following up your case in order to verify if you still need assistance on this thread. If so, please do not hesitate and let me know and I will be more than happy to assist you. Regards, Emeth O. Intel Customer Support Technician Under Contract to Intel Corporation

[Emeth_O_Intel]

Hi Naoya,

Due to the fact that I have not seen any activy on this thread, I will proceed and close it.

Please if you have any other question, do not hesitate and contact us back and we will be more than happy to assist you.

Regards,

Emeth O. 

Intel Customer Support Technician

Under Contract to Intel Corporation