[EXTERNAL] Re: [RFC PATCH 3/3] dts: add cryptodev testsuite

[Akhil Goyal]

IMO, it varies from platform to platform. Adding Kai Ji for details on QAT.
However, you can use as many VFs as you want. Generally, we use 1VF per core.

Do you plan to include other platforms as well?
How about openssl PMD?

From: Andrew Bailey <abailey at iol.unh.edu>
Sent: Wednesday, January 21, 2026 9:57 PM
To: Akhil Goyal <gakhil at marvell.com>
Cc: Patrick Robb <probb at iol.unh.edu>; fanzhang.oss at gmail.com; luca.vizzarro at arm.com; dmarx at iol.unh.edu; dev at dpdk.org
Subject: Re: [EXTERNAL] Re: [RFC PATCH 3/3] dts: add cryptodev testsuite

have a few more questions for you regarding testing cryptodev. As a reminder, we're developing a test suite to detect performance or functional regressions in cryptodev, using the dpdk-test-crypto-perf application with Throughput, Latency,
ZjQcmQRYFpfptBannerStart
Prioritize security for external emails:
Confirm sender and content safety before clicking links or opening attachments
    Report Suspicious  <https://us-phishalarm-ewt.proofpoint.com/EWT/v1/CRVmXkqW!uK3X-9Eb4hoauMV9EzXzTis813o0RIZqUEPdy9lsu2a7UusF9ZitMxmorqa-v2d3QcZ2mvqb8d1cVEIgIEXxH0zaLC9PUtWvU6Br3tNjRVRhEmxF3rRTk9zVNhM$>   ‌
ZjQcmQRYFpfptBannerEnd
 have a few more questions for you regarding testing cryptodev. As a reminder, we're developing a test suite to detect performance or functional regressions in cryptodev, using the dpdk-test-crypto-perf application with Throughput, Latency, and Verify ptest options. Right now we are testing with an Intel QAT 8970. It has 3 physical functions, and DTS automatically creates 16 VFs on each PF when the testrun starts. I have some questions regarding which VFs to use and how many to use when checking for regressions:

  *   Is it appropriate to specify only a single virtual function in the allowed ports list for our test suite?
  *   Under what circumstances should we allow multiple virtual functions?
  *   When allowing multiple virtual functions, should we include one from each physical function?

Thank you for your guidance.

On Mon, Jan 12, 2026 at 2:20 PM Akhil Goyal <gakhil at marvell.com<mailto:gakhil at marvell.com>> wrote:
Hi Andrew,

  1.  Which metrics in the attached table below are relevant for checking for performance regressions in throughput testing? MOps? Gbps?
Mops and Gbps both are calculation based(Mops * packet_size = Gbps). You can use anyone of them.

  1.  What would be an acceptable throughput deviation between runs, and what percentage deviation should be considered a failure? For instance, right now I am thinking that if the test returns a Gbps reduction of 2% or greater from the given baseline, that will trigger a failure.
IMO, 5% should be good enough.

Regards,
Akhil

From: Andrew Bailey <abailey at iol.unh.edu<mailto:abailey at iol.unh.edu>>
Sent: Monday, January 12, 2026 11:28 PM
To: Patrick Robb <probb at iol.unh.edu<mailto:probb at iol.unh.edu>>; Akhil Goyal <gakhil at marvell.com<mailto:gakhil at marvell.com>>; fanzhang.oss at gmail.com<mailto:fanzhang.oss at gmail.com>
Cc: luca.vizzarro at arm.com<mailto:luca.vizzarro at arm.com>; dmarx at iol.unh.edu<mailto:dmarx at iol.unh.edu>; dev at dpdk.org<mailto:dev at dpdk.org>
Subject: [EXTERNAL] Re: [RFC PATCH 3/3] dts: add cryptodev testsuite

Hello, I am Andrew, a software test developer at UNH-IOL working on the DPDK Test Suite (DTS) component in DPDK. We're adding crypto device tests (both functional and performance) to DTS using the dpdk-test-crypto-perf application. These
ZjQcmQRYFpfptBannerStart
Prioritize security for external emails:
Confirm sender and content safety before clicking links or opening attachments
    Report Suspicious  <https://us-phishalarm-ewt.proofpoint.com/EWT/v1/CRVmXkqW!uK3X-9Eb4hoauMMx8776juphaxuhWSpVPz0jRFbXZsENdFzDoFkHkisIFSRgpqYCKAsvJZ0J93ubWr88-ERDphsvU_UUPsR6-5g8RuDP-gdOzh7iwbOo1NwbQlg$>   ‌
ZjQcmQRYFpfptBannerEnd
Hello,

I am Andrew, a software test developer at UNH-IOL working on the DPDK Test Suite (DTS) component in DPDK. We're adding crypto device tests (both functional and performance) to DTS using the dpdk-test-crypto-perf application. These are tests which can be run on real hardware, I.e. We have an Intel 8970 crypto card at the lab we are using for testing. I'm currently developing a performance test suite using option "--ptest throughput" that will fail tests when metrics deviate by n% from a baseline. I would like to get your feedback regarding how I can correctly check for performance regressions in the test cases by using dpdk-crypto-perf.

1. Which metrics in the attached table below are relevant for checking for performance regressions in throughput testing? MOps? Gbps?
2. What would be an acceptable throughput deviation between runs, and what percentage deviation should be considered a failure? For instance, right now I am thinking that if the test returns a Gbps reduction of 2% or greater from the given baseline, that will trigger a failure.

[cid:image001.png at 01DC8B3A.FC2EC5C0]
Any guidance you can provide would be greatly appreciated.
Best regards,
Andrew
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mails.dpdk.org/archives/dev/attachments/20260121/80457649/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.png
Type: image/png
Size: 122934 bytes
Desc: image001.png
URL: <http://mails.dpdk.org/archives/dev/attachments/20260121/80457649/attachment-0001.png>