That's not really true.  Remember that all the algorithms are implemented in 
hardware inside the CEX cards.  The hardware is so fast that all the algorithms 
are essentially the same speed, relative to the rest of the time which is spent 
in software and communications.  If you look at the performance numbers in the 
document Greg Boyd referenced, you'll see that triple-DES (TDES) is actually 
FASTER than AES in this system.  Where you would expect AES to be faster is 
when the algorithms are implemented in software, and where there is no 
additional overhead like communications to a separate device.

Here are some of the numbers from that performance document.  For each 
algorithm, it shows performance in kbytes/second from a host application 
program, for three different sizes of input data that you are encrypting.  
(This table probably gets messed up depending on whether you are displaying in 
fixed or proportional fonts - sorry about that.)

Algorithm      64 bytes       64K bytes       1M bytes
------------      -----------       ------------        -----------
TDES                161               13372            14000
AES-256           149                 9645              9995
AES-128           148                 9822            10191

I'm not sure why the AES performance is lower on System z than TDES, since 
performance numbers from a recent version of the card firmware running on 
System x Linux show the opposite - here are those numbers, where I have 
separate data for AES using the older fixed-length key token structures and 
using the newer variable-length key tokens.

Algorithm                    64 bytes       64K bytes       1M bytes
------------                   -----------       ------------        -----------
TDES                               199             53202            68770
AES-128 (fixed)               209             60277            73840
AES-128 (variable)          206             57546            73090
AES-256 (fixed)               211             61315            71790
AES-256 (variable)          203             57361            72840

So, the same card running under Linux in a System x server does indeed have AES 
numbers greater than TDES numbers - but by a very small amount.  For those who 
are curious, I suspect the large-block performance on System x is so much 
higher because that system allows us to send more data to the crypto card at a 
time - meaning there are fewer calls to the card than on System z, where it is 
chopped into smaller packets.

Todd Arnold
IBM Cryptographic Coprocessor development

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to