Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 15 Oct 2024 17:03:45 +0000
From:      bugzilla-noreply@freebsd.org
To:        bugs@FreeBSD.org
Subject:   [Bug 282115] GELI + QAT - 8955 performance is less than software encoding for block device acceleration
Message-ID:  <bug-282115-227@https.bugs.freebsd.org/bugzilla/>

next in thread | raw e-mail | index | archive | help
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=3D282115

            Bug ID: 282115
           Summary: GELI + QAT - 8955 performance is less than software
                    encoding for block device acceleration
           Product: Base System
           Version: 14.1-STABLE
          Hardware: amd64
                OS: Any
            Status: New
          Severity: Affects Some People
          Priority: ---
         Component: kern
          Assignee: bugs@FreeBSD.org
          Reporter: james@elstone.net

Created attachment 254268
  --> https://bugs.freebsd.org/bugzilla/attachment.cgi?id=3D254268&action=
=3Dedit
Memory Disk QAT + Geli Test Steps

Having installed an Intel QAT 8955 card and tested performance, there is a
significant decrease in speed with the QAT card enabled compared to software
crypto.=20

The CPU used for these offline tests was an old Intel(R) Xeon(R) CPU E5-240=
3 0
@ 1.80GHz that has the AES-NI instruction set. Fresh install of FreeBSD 14.=
1-p5
was used for each test, that than a custom kernel with AES-NI removed.

The disk used to test against in the results below is a WD BLACK SN750 SE i=
n a
PCI slot adapter. The DD speed to a mounted volume for this disk is 1532 MB=
/s.

The script to generate these results is attached for referencem but uses a
memory disk for testing maximal throughput.

The /boot/loader.conf is:
 dev.qat.0.cfg_services =3D "sym;asym"
 qat_dh895xcc_fw_load=3D"YES"
 qat_load=3D"YES"
 cryptodev_load=3D"YES"

Parameters: A:HMAC/SHA256 E:AES-CBC L:256
Crypto: hardware
irq85: qat0:b0                  77310798       1368
irq86: qat0:b1                 120404887       2130
DD - DISK_BLK_SIZE:4096 DD_BLK_SIZE:8388608 DD_COUNT:25
** write speed: (66,475,964 bytes/sec)
irq85: qat0:b0                  77310798       1368
irq86: qat0:b1                 120833699       2137

Parameters: A:HMAC/SHA256 E:AES-XTS L:256 BS:4096
Crypto: hardware
irq85: qat0:b0                  77310798       1367
irq86: qat0:b1                 121389848       2147
DD - DISK_BLK_SIZE:4096 DD_BLK_SIZE:8388608 DD_COUNT:25
** Write speed: (61,894,122 bytes/sec)
irq85: qat0:b0                  77310798       1367
irq86: qat0:b1                 121841632       2155


Parameters: A:HMAC/SHA256 E:AES-CBC L:256 BS:4096
Crypto: software
DD - DISK_BLK_SIZE:4096 DD_BLK_SIZE:8388608 DD_COUNT:25
** write speed: (110,486,470 bytes/sec)

Parameters: A:HMAC/SHA256 E:AES-XTS L:256
Crypto: software
DD - DISK_BLK_SIZE:4096 DD_BLK_SIZE:8388608 DD_COUNT:25
** write speed: (102,362,135 bytes/sec)

The above shows a drop in performance of 40%.

Additionally when using the attached memory disk example, the throughput of
this system tops out at ~135Mbps when writing to a mounted encrypted ramdisk
UFS volume, with the 4 cores operating at about 25% each.

--=20
You are receiving this mail because:
You are the assignee for the bug.=



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?bug-282115-227>