OFS : Data Path of QSFP from HE_HSSI

[attdoner]

Hi Team,

I am using N6001 with Intel OFS (OFS PCIe Attach) . I wanted to test data transfer from Host1 to Host2 as below.
Host1 -> N6001(Port0) -> OtherNIC(Port0) -> Host2.
Is there any sample application available in OPAE/OFS which performs this transaction? Can you please suggest.

In addition to above query, I wants to know how HE-HSSI sends data to QSFP?. The hssi application shows "No eth interface, so not honoring --eth-loopback". ( https://ofs.github.io/ofs-2024.1-1/hw/n6001/user_guides/ug_qs_ofs_n6001/ug_qs_ofs_n6001/  Section 7. Send traffic through the 10G AFU). Does this mean the HE_HSSI will not communicate with E-Tile or QSFPs (FIM) ?

Do I need to enable any parameters to pass over QSFPs?

Thankyou.

[attdoner]

Hi @JohnT_Intel ,
In addition to above queries, the Transfer rate shows throughput nearly 25Gbps for 10G using 'sudo hssi hssi_10g  --num-packets 1 '. But when I execute for 100G, I could see error message. Can I know what causes this issue? How to resolve this error?

$ sudo hssi hssi_100g --num-packets 10

[hssi_100g] [error] no accelerator found with id: 43425XX6-9XX2-3XX2-bXXa-aXXX4b53XXX2

[JohnT_Intel]

Hi,

Are you running " sudo hssistats --clear" to clear the stat? I am able to clear the stat from my side without any issue.

May I know which image are you using for your board? I suspect that you are using 10G FIM which is why it is not able to detect the AFU GUID for it to run the test. The provided image is 2x4x25GbE.

You will need to change the FIM design to 100G design

[attdoner]

Hi @JohnT_Intel ,

I have tried " sudo hssistats --clear" still the values used to be there.

We are using https://github.com/otcshare/intel-ofs-fim/releases/tag/ofs-4.1.1. 
Yes, Ethernet configurations: 2x4x25GbE.

* Whether the FIM in this link is 10G ?

* Do you use other image for 2x4x25GbE?
* Can we know is there any FIM with 100G ?

[JohnT_Intel]

Whether the FIM in this link is 10G ?

* Do you use other image for 2x4x25GbE?

Currently the official release image is 2x4x25GbE.

* Can we know is there any FIM with 100G ?

I am checking with engineering to see if we have this image.

[attdoner]

Hi @JohnT_Intel ,
  
    Thanks for the update.
    I tried with this FIM (https://github.com/otcshare/intel-ofs-fim/releases/download/ofs-3.2.1/FIM_x8_2x1x100G.zip ). Even the error is same.

[user@user ~]$ sudo fpgainfo phy
Intel Acceleration Development Platform N6001
Board Management Controller NIOS FW version: 3.15.1 
Board Management Controller Build version: 3.15.1 
//****** PHY ******//
Interface                        : DFL
Object Id                        : 0xF000000
PCIe s:b:d.f                     : 0000:01:00.0
Vendor Id                        : 0x8086
Device Id                        : 0xBCCE
SubVendor Id                     : 0x8086
SubDevice Id                     : 0x1770
Socket Id                        : 0x00
Ports Num                        : 01
Bitstream Id                     : 0x501060255341BA8
Bitstream Version                : 5.0.1
Pr Interface Id                  : 98c2df0a-7c39-5bda-9d5b-6a38550cd3d8
QSFP0                            : Not Connected 
QSFP1                            : Not Connected 
//****** HSSI information ******//
HSSI version                     : 1.0  
Number of ports                  : 4  
Port0                            :100GCAUI-4   UP
Port4                            :100GCAUI-4   UP
Port8                            :100GCAUI-4   DOWN
Port12                           :100GCAUI-4   DOWN

[user@user ~]$ sudo hssi hssi_100g
[2024-11-25 11:06:06.438] [hssi_100g] [error] no accelerator found with id: 43425ee6-92b2-4742-b03a-bd8d4a533812

[JohnT_Intel]

Hi,

The reason is that particular FIM is having different Accelerator GUID. I am currently checking to see if we have example design for this testing.

[attdoner]

Hi @JohnT_Intel ,
Is there any update on example design ?
In addition to my query I have modified few changes in N6001 AFU, but I am facing mbox_write timeout as below

sudo hssi hssi_100g

100G loopback test
  port: 0
  eth_loopback: on
  num_packets: 1
  gap: none
  src_address: 11:22:33:44:55:66
    (bits): 0x112233445566
  dest_address: 77:88:99:aa:bb:cc
    (bits): 0x778899aabbcc
  pattern: random
  start size: 40
  end size: 2580
  end select: pkt_num
  eth: 
  continuous mode: off
  monitor duration: 0 sec

No eth interface, so not honoring --eth-loopback. Try using the hssiloopback command instead.
mbox_write timed out [a]

[JohnT_Intel]

Hi,

Unfortunately we do not have the design for the 100G. You will need to build the FIM with this command "./ofs-common/scripts/common/syn/build_top.sh --ofss tools/ofss_config/hssi/hssi_2x100.ofss n6001 work_n6001"

Thanks.

[attdoner]

Hi @JohnT_Intel ,
Thanks for the update, I will compile and get back to you.

Meanwhile, with 10g_hssi tests, I could see throughput difference.

TX Packet Count (N6001)	TX Throughput(N6001) GB/s
1	25.7812
50	16.0559
500	15.5692
5000	15.5638
50000	15.5708
500000	15.5693
5000000	15.5692
9586980	15.5652
9586981	6.96E-07
10000000	0.643037
1000000000	0.177615

As packet count increases, the throughput decreases. In above table,
For Packet Count 1: The TX throughput = 25Gbps
For Packet Count 50: The TX throughput = 16.05Gbps

Can I know why the "throughput" values are lesser for higher packet counts?

[JohnT_Intel]

Hi,

It will be limited due to traffic generator. When you have too much packet to sent, it is impacting the traffic generator to generate the packet needed.

[attdoner]

Hi @JohnT_Intel ,

With your previous reply "
Unfortunately we do not have the design for the 100G. You will need to build the FIM with this command "./ofs-common/scripts/common/syn/build_top.sh --ofss tools/ofss_config/hssi/hssi_2x100.ofss n6001 work_n6001"
"

I had tried this command , and I face the below issue.

sudo hssi hssi_100g

100G loopback test
  port: 0
  eth_loopback: on
  num_packets: 1
  gap: none
  src_address: 11:22:33:44:55:66
    (bits): 0x112233445566
  dest_address: 77:88:99:aa:bb:cc
    (bits): 0x778899aabbcc
  pattern: random
  start size: 40
  end size: 2580
  end select: pkt_num
  eth: 
  continuous mode: off
  monitor duration: 0 sec

No eth interface, so not honoring --eth-loopback. Try using the hssiloopback command instead.
mbox_write timed out [a]

[JohnT_Intel]

Hi,

Can you share with me your bitstream?

[attdoner]

Hi @JohnT_Intel ,
I had attahced the bitstream generated using this command "./ofs-common/scripts/common/syn/build_top.sh --ofss tools/ofss_config/hssi/hssi_2x100.ofss n6001 work_n6001"

https://drive.google.com/file/d/1Vz35lGX3VPandzAgsU8NGwGR-9Z6ItOw/view?usp=sharing

Please let us know any update on this

[JohnT_Intel]

Hi,

I observed that you are sharing .pof file. Can you share .bin file that I can directly load into the board?

[attdoner]

Hi @JohnT_Intel ,

The .bin files : https://drive.google.com/file/d/1PdiDg2SX70VpYdDg1G5l46wRGfkmc9a5/view?usp=sharing

[omprasad]

Subscribing for the same

[JohnT_Intel]

Hi,

I have try the bin file that you provided and I do not observed the mbox error that you mention.

[attdoner]

Hi @JohnT_Intel ,
Can you please share the output of 100g using below command
sudo hssi hssi_100g

[JohnT_Intel]

Hi,

Below is what I observed.

[attdoner]

Hi @JohnT_Intel ,
Can you please reshare the image.
The image is not able to view.

[JohnT_Intel]

Hi,

Attach is the screenshot