Hello,

We are going to go ahead and enable this testing per-patch starting today, but again we looked at the performance metrics and they seemed reasonable - so we should be good to go.

If you do end up wanting to take a closer look, let us know what info we can provide. We can also grant you VPN/SSH access to the DUT system if you want to take a look for any reason. 

Thanks.

On Wed, Apr 23, 2025 at 10:58 AM Cody Cheng <ccheng@iol.unh.edu> wrote:
Hi Kai,

I am reaching out to you regarding our efforts to set up a new QAT
device at the DPDK Community Lab. This is for adding a new QAT Device,
beyond the QAT PCI card that we set up and began testing last year for
an ARM Ampere System.

I've set up testing for the Intel QAT 4xxx series device, but I just
wanted to double-check that the performance metrics we are seeing from
DPDK's test-crypto-perf application look acceptable to you. We are
running a DTS testsuite which uses the dpdk-test-crypto-perf
application in throughput, latency, and verify mode for testing. If
you want to see the test plan, you can read it here:
https://git.dpdk.org/tools/dts/tree/test_plans/crypto_perf_cryptodev_perf_test_plan.rst

Could you review the test results and let me know if they seem correct
for this QAT SKU?

Here are the test results. The metrics you are looking for should be
visible in the perf_cryptodev_result.json file:

https://drive.google.com/drive/folders/1PfFziVYrD_Z2HQfXprty2lLNq5YX1Y7Y

Thanks,
Cody Cheng


On Tue, Apr 8, 2025 at 2:36 PM Mcnamara, John <john.mcnamara@intel.com> wrote:
>
> Hi,
>
>
>
> You can bounce the numbers, or any questions, off Kai Ji (added).
>
>
>
> John
>
>
>
>
>
> From: Patrick Robb <probb@iol.unh.edu>
> Sent: Tuesday, April 8, 2025 6:03 PM
> To: Richardson, Bruce <bruce.richardson@intel.com>
> Cc: Cody Cheng <ccheng@iol.unh.edu>; Dean Marx <dmarx@iol.unh.edu>; Adam Hassick <ahassick@iol.unh.edu>; Nicholas Pratte <npratte@iol.unh.edu>; ci@dpdk.org
> Subject: Intel QuickAssist crypto operations testing
>
>
>
> Hi Bruce,
>
>
>
> Do you remember last year when we had a tech board call about the server refresh at UNH-IOL, you recommended that for the Intel processor DUT system, we get a processor SKU which includes a QAT device? We did this, and the SKU we selected was Intel(R) Xeon(R) Gold 5415+.
>
>
>
> Those devices are QAT 4xxx series QAT 4942 as reported by dpdk-devbind.py --status:
>
>
>
> Crypto devices using kernel driver
> ==================================
> 0000:76:00.0 '4xxx Series QAT 4942' numa_node=0 drv=4xxx unused=qat_4xxx,vfio-pci
> 0000:f3:00.0 '4xxx Series QAT 4942' numa_node=1 drv=4xxx unused=qat_4xxx,vfio-pci
>
>
>
> Cody has been setting up the testing for this QAT device, which is done via DPDK's test-crypto-perf application. He is satisfied with the metrics the card is getting with DPDK (he compared them against some reports Intel has published and also what we see on a QAT PCI card on an Ampere ARM system here at UNH).
>
>
>
> So, he thinks we are ready to go online. However, in order to be careful, we were wondering if we should run the information we have by an Intel contact first. Is there any DPDK/quickassist/crypto person on your team who could serve as a contact for this purpose? If so, I think Cody can add them to this thread and share the more detailed info for that person for them to approve (or disapprove).
>
>
>
> Thanks let me know!
>
>
>
> -Patrick
>
>
>
>