VPP show buffers issue on ls1046ardb

[skyler]

Hey, Guys,

I have started VPP on my ls1046ardb board, and did the test cases successfully from lsdk user guide.

Now, I wanna investigate more about performance, so I checked the buffer usage on VPP, but the statistics are very confusing, though the function is all right, packets forwarding right:

vpp# show buffers
Pool Name Index NUMA Size Data Size Total Avail Cached Used
default-numa-0 0 0 2496 2048 16000 15968 34 42949672

vpp# show buffers
Pool Name Index NUMA Size Data Size Total Avail Cached Used
default-numa-0 0 0 2496 2048 8000 8000 0 0

For comparison, I tested the same case on the X86 platform<VPP 21.01>, I can see the "show buffer" works well:

vpp# show buffers
Pool Name Index NUMA Size Data Size Total Avail Cached Used
default-numa-0 0 0 2304 2048 16808 16296 152 360
vpp# show buffers
Pool Name Index NUMA Size Data Size Total Avail Cached Used
default-numa-0 0 0 2304 2048 16808 16296 154 358

So there are two questions:

1, Is the VPP repo in lsdk21.08 the same as https://github.com/nxp-qoriq/vpp/releases/tag/21.08-LSDK and modified from the origin repo https://github.com/FDio/vpp/releases/tag/v21.01 

2. do you guys have any idea about this issue, any relative modifications about buffer pool management you have done?

Any replies will be helpful.

Thanks

[Bio_TICFSL]

Hello,

Please refer to NXP latest LSDK release for Layerscape https://www.nxp.com.cn/docs/en/user-guide/LLDPUG_RevL6.1.22_2.0.0.pdf

VPP version is as the following.

VPP v23.02 build with DPDK 22.11.

LLDPUG is based on Yocto, LSDK is based on flex-builder.

LLDPUG is newer than LSDK.

LLDPUG will replace LSDK in feature.

Regards

[Bio_TICFSL]

Please refer to NXP latest LSDK release for Layerscape https://www.nxp.com.cn/docs/en/user-guide/LLDPUG_RevL6.1.22_2.0.0.pdf

VPP version is as the following.

VPP v23.02 build with DPDK 22.11.

LLDPUG is based on Yocto, LSDK is based on flex-builder.

LLDPUG is newer than LSDK.

LLDPUG will replace LSDK in feature.

Regards

[Bio_TICFSL]

LLDPUG is based on Yocto, LSDK is based on flex-builder.

LLDPUG is newer than LSDK.

LLDPUG will replace LSDK in feature.

Regards

[skyler]

hey Bio_TICFSL,

Thanks for the advice, I have booted the latest LSDK release image, and the buffers seem to show right.

But there is another issue, I'm doing the forwarding test by VPP.

Topology:

PC<port0> <-----> <fm1-mac3> ls1046ardb <fm1-mac4> <------> PC<port1>

Result:

PC Statistics:

FM Port:

VPP:

According to the result, you can see the FM port got all the packets, but VPP dpdk-input rx lost  about 1.5M packets, and I can't see anything wrong in vpp.

Do you have any idea about where to check the lost packets?

Best Regards

Skyler.

[skyler]

Hey, Bio_TICFSL,

A few updates:

1) the lost packets is located, all in port_discard_frame </sys/devices/platform/soc/1a00000.fman/1a8a000.port/statistics>

2) I tested on rgmii port with trex<100% bandwidth>, when frame size is 1518, no frame discard, but once frame size is 64, many discards.

As you know, a smaller frame size means more frames. so maybe the frame queue is up to threshold?

I'm still working on it.

Any suggestion will be helpful.

Best Regards,

Skyler.