On Sat, Nov 04, 2006 at 11:09:56PM -0500, Greg Mortensen wrote: > Darrin Chandler <[EMAIL PROTECTED]> wrote: > > >For desktop/server use, hardware acceleration for crypto seems > >increasingly irrelevant as processors become faster. Yawn. > > From a VIA PadlockACE equipped SBC: > > 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes > aes-128-cbc 31885.24k 118568.67k 312349.58k 535048.83k 649099.91k > > From a "irrelevant as processors become faster" i386: > > aes-128-cbc 61905.43k 83868.59k 91948.85k 93908.47k 93081.82k > > Yawn indeed. > > >For "appliances" such as soekris, WRAP, et al, crypto in hardware can > >still be quite important. > > It is here that's it's quite irrelevant, as these little CPUs cannot > feed the accelerator fast enough. From a vpn1411 equipped Soekris: > > aes-128-cbc 63.65k 250.39k 944.13k 2953.46k 7989.79k > > >I see the occasional post here about someone trying to make sure their > >accelerator is being used in OpenBSD ... > > ...because the userland -> kernel -> userland transition makes a > hardware accelerated Soekris slower than a stock one, except for large > block sizes. From a net4801: > > aes-128-cbc 2408.59k 2738.99k 2810.27k 2841.62k 2766.26k
Thanks for posting some hard numbers! Very interesting and good to know. Can you say what the "irrelevant" i386 machine is? Lots of difference between a 90MHz PentiumI and a 3GHz Opteron, and I'd like to know where those numbers fit in. -- Darrin Chandler | Phoenix BSD Users Group [EMAIL PROTECTED] | http://bsd.phoenix.az.us/ http://www.stilyagin.com/ |
