intel / intel/QAT_Engine

Chained cipher performance on Denverton

Open
#145 2 comments 0 reactions 0 assignees View on GitHub
Dominant language
C
Stars
443
Forks
137
PR merge metrics
No merged PRs in 30d

Description

My Rangeley CPU finally died on me last week and so I upgraded to a Denverton.

While I was already aware on Rangeley that offloading symmetric ciphers with buffer sizes any less than 4k resulted in significantly poorer performance than just using the CPU instructions and indeed prepared to see the same thing with Denverton, I was quite perplexed to see the results below.

QAT is working fine, as the results from the RSA speed test show. However, even with large buffers, the speed for chained ciphers on Denverton is way slower than not offloading.

Is this normal?

RSA results: more or less what I expected

```
$ openssl speed -engine qat -elapsed -async_jobs 72 rsa2048

sign verify sign/s verify/s
rsa 2048 bits 0.000105s 0.000013s 9486.0 74558.1

$ openssl speed -elapsed -async_jobs 72 rsa2048

sign verify sign/s verify/s
rsa 2048 bits 0.002705s 0.000078s 369.7 12815.9
```

AES-CBC: not at all what I expected. The results are the same even when the number of async jobs are upped to 72

```
$ openssl speed -engine qat -elapsed -async_jobs 32 -multi 2 -evp aes-128-cbc-hmac-sha1

evp 180743.19k 470527.89k 782870.78k 971217.58k 906024.28k 893157.38k

$ openssl speed -elapsed -async_jobs 32 -multi 2 -evp aes-128-cbc-hmac-sha1

evp 318426.91k 674715.48k 897365.59k 1002786.47k 1050793.30k 1052153.17k
```

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.