Skip to contentSkip to conversation

REQUIRED CIPHERS TEST takes 20 minutes each for aes-related ciphers

dm-crypt [email protected] 3 participants 6 messages Nov 28, 2013

lore
Share & tools
lore
#0 lore raw

Hello,

I just compiled cryptsetup-1.6.2 on a machine with AES-NI. During the compilation there were some tests. The lines of the output when the issue occurred were:

REQUIRED CIPHERS TEST
# Algorithm | Key | Encryption | Decryption

     aes-cbc   256b   975.0 MiB/s  2037065.0 MiB/s
     aes-lrw   384b  103909.0 MiB/s  158691.0 MiB/s
     aes-xts   512b  1731833.0 MiB/s  1681613.0 MiB/s
 twofish-cbc   256b   241.3 MiB/s   499.5 MiB/s
 twofish-lrw   384b   448.6 MiB/s   453.0 MiB/s
 twofish-xts   512b   497.0 MiB/s   496.0 MiB/s
 serpent-cbc   256b   108.5 MiB/s   351.3 MiB/s
 serpent-lrw   384b   333.7 MiB/s   333.0 MiB/s
 serpent-xts   512b   363.3 MiB/s   341.0 MiB/s
blowfish-cbc   256b   129.6 MiB/s   500.5 MiB/s
des3_ede-cbc   192b    34.3 MiB/s    35.4 MiB/s
   cast5-cbc   128b   121.9 MiB/s   474.0 MiB/s

The tests for aes-cbc, aes-lrw and aes-xts took very long (nearly one hour for all three combined) on a Intel Core i5-4670K. All other tests were finished in a few seconds.

Is this wanted behavior?

Best regards,
Christian Weinz

#1 lore raw

[email protected]:

Quoted 8 lines

Hello,

I just compiled cryptsetup-1.6.2 on a machine with AES-NI. During the compilation there were some tests. The lines of the output when the issue occurred were:

REQUIRED CIPHERS TEST
# Algorithm | Key | Encryption | Decryption

Quoted 12 lines
     aes-cbc   256b   975.0 MiB/s  2037065.0 MiB/s
     aes-lrw   384b  103909.0 MiB/s  158691.0 MiB/s
     aes-xts   512b  1731833.0 MiB/s  1681613.0 MiB/s
 twofish-cbc   256b   241.3 MiB/s   499.5 MiB/s
 twofish-lrw   384b   448.6 MiB/s   453.0 MiB/s
 twofish-xts   512b   497.0 MiB/s   496.0 MiB/s
 serpent-cbc   256b   108.5 MiB/s   351.3 MiB/s
 serpent-lrw   384b   333.7 MiB/s   333.0 MiB/s
 serpent-xts   512b   363.3 MiB/s   341.0 MiB/s
blowfish-cbc   256b   129.6 MiB/s   500.5 MiB/s
des3_ede-cbc   192b    34.3 MiB/s    35.4 MiB/s
   cast5-cbc   128b   121.9 MiB/s   474.0 MiB/s

when these tests ran did all cores process or just 1 single core ?

do you know if cryptsetup can take advantage of multi threading on multi core processors ?

Quoted 14 lines

The tests for aes-cbc, aes-lrw and aes-xts took very long (nearly one hour for all three combined) on a Intel Core i5-4670K. All other tests were finished in a few seconds.

Is this wanted behavior?

Best regards,
Christian Weinz
_______________________________________________
dm-crypt mailing list
[email protected]
http://www.saout.de/mailman/listinfo/dm-crypt

#2 lore raw
Quoted 22 lines · El 2013-11-28 14:30, [email protected] escribió:

[email protected]:

Quoted 8 lines

Hello,

I just compiled cryptsetup-1.6.2 on a machine with AES-NI. During the compilation there were some tests. The lines of the output when the issue occurred were:

REQUIRED CIPHERS TEST
# Algorithm | Key | Encryption | Decryption

Quoted 12 lines
     aes-cbc   256b   975.0 MiB/s  2037065.0 MiB/s
     aes-lrw   384b  103909.0 MiB/s  158691.0 MiB/s
     aes-xts   512b  1731833.0 MiB/s  1681613.0 MiB/s
 twofish-cbc   256b   241.3 MiB/s   499.5 MiB/s
 twofish-lrw   384b   448.6 MiB/s   453.0 MiB/s
 twofish-xts   512b   497.0 MiB/s   496.0 MiB/s
 serpent-cbc   256b   108.5 MiB/s   351.3 MiB/s
 serpent-lrw   384b   333.7 MiB/s   333.0 MiB/s
 serpent-xts   512b   363.3 MiB/s   341.0 MiB/s
blowfish-cbc   256b   129.6 MiB/s   500.5 MiB/s
des3_ede-cbc   192b    34.3 MiB/s    35.4 MiB/s
   cast5-cbc   128b   121.9 MiB/s   474.0 MiB/s
Quoted 1 line

when these tests ran did all cores process or just 1 single core ?

For aes-cbc, aes-lrw and aes-xts just one core was used (and maxed out). However, it took orders of magnitudes longer.

Quoted 2 lines

do you know if cryptsetup can take advantage of multi threading on multi core processors ?

No.

Quoted 14 lines
Quoted 14 lines

The tests for aes-cbc, aes-lrw and aes-xts took very long (nearly one hour for all three combined) on a Intel Core i5-4670K. All other tests were finished in a few seconds.

Is this wanted behavior?

Best regards,
Christian Weinz
_______________________________________________
dm-crypt mailing list
[email protected]
http://www.saout.de/mailman/listinfo/dm-crypt

#3 lore raw
In reply to [email protected]
Quoted 4 lines · On 11/28/2013 03:30 PM, [email protected] wrote:

when these tests ran did all cores process or just 1 single core ?

do you know if cryptsetup can take advantage of multi threading on multi core processors ?

FYI this test doesn't use dmcrypt at all, only kernel crypto API, so it depends how crypto layer module processes it.

Anyway, it doesn't matter here - the whole test basically runs encryption of 1MB buffer over userspace API to kernel crypto layer.

It cannot take so long - it is clear problem somewhere.

Milan

#4 lore raw
In reply to [email protected]
Quoted 11 lines · On 11/28/2013 02:25 PM, [email protected] wrote:

Hello,

I just compiled cryptsetup-1.6.2 on a machine with AES-NI. During the compilation there were some tests. The lines of the output when the issue occurred were:

REQUIRED CIPHERS TEST
# Algorithm | Key | Encryption | Decryption

      aes-cbc   256b   975.0 MiB/s  2037065.0 MiB/s
      aes-lrw   384b  103909.0 MiB/s  158691.0 MiB/s
      aes-xts   512b  1731833.0 MiB/s  1681613.0 MiB/s

Wow :) This is apparently bug somewhere.

Can you please sent me (or better create new issue
http://code.google.com/p/cryptsetup/issues/entry)

Please attach cryptsetup benchmark with --debug,
lsmod, cat /proc/crypto and uname -a.
(I will probably need more info later though...)

Thanks!
Milan

#5 lore raw
Quoted 13 lines · Milan Broz
Quoted 11 lines · On 11/28/2013 02:25 PM, [email protected] wrote:

Hello,

I just compiled cryptsetup-1.6.2 on a machine with AES-NI. During the compilation there were some tests. The lines of the output when the issue occurred were:

REQUIRED CIPHERS TEST
# Algorithm | Key | Encryption | Decryption

      aes-cbc   256b   975.0 MiB/s  2037065.0 MiB/s
      aes-lrw   384b  103909.0 MiB/s  158691.0 MiB/s
      aes-xts   512b  1731833.0 MiB/s  1681613.0 MiB/s
Quoted 8 lines

Wow :) This is apparently bug somewhere.

Can you please sent me (or better create new issue
http://code.google.com/p/cryptsetup/issues/entry)

Please attach cryptsetup benchmark with --debug,
lsmod, cat /proc/crypto and uname -a.
(I will probably need more info later though...)

Here you go: http://code.google.com/p/cryptsetup/issues/detail?id=186

Thanks for looking into this.

Quoted 2 lines

Thanks!
Milan