On Thu, Mar 09, 2006 at 12:51:51AM +0000, Stuart Henderson wrote: > On 2006/03/08 18:29, Gordon Grieder wrote: > > The speed with and without userland crypto enabled was the same > > Doesn't necessarily mean anything - overheads might be doing this.
See below. > The clearest indication is probably to watch the hifn's irq in > systat vmstat. > > > for things the cards support (ie.: sha1, rsa, et al). > > rsa?: "The 7951, 7955 and 7956 have support for public key operations > which are not yet supported." Right, my mistake. <duh> Figured something else out, it's working with SSH fine. Anything I do remotely hikes up the interrupts on the hifn0 device. (ie.: "find /" throws the interrupt count high as the stuff scrolls by) It seems that it's openssl that isn't working. That's progress. :) > sha1 should be supported, though how are you testing it? > how about 3des or aes-cbc?? Just with a script which disables userland crypto, runs a bench, enables it and runs the same bench. [snip] sudo sh bench.sh sha1 Password: kern.usercrypto: 1 -> 0 kern.userasymcrypto: 1 -> 0 To get the most accurate results, try to run this program when this computer is idle. Doing sha1 for 3s on 16 size blocks: 570605 sha1's in 3.00s Doing sha1 for 3s on 64 size blocks: 485408 sha1's in 3.01s Doing sha1 for 3s on 256 size blocks: 326027 sha1's in 3.00s Doing sha1 for 3s on 1024 size blocks: 140790 sha1's in 2.97s Doing sha1 for 3s on 8192 size blocks: 22362 sha1's in 2.97s kern.usercrypto: 0 -> 1 kern.userasymcrypto: 0 -> 1 To get the most accurate results, try to run this program when this computer is idle. Doing sha1 for 3s on 16 size blocks: 567198 sha1's in 2.98s Doing sha1 for 3s on 64 size blocks: 481930 sha1's in 2.98s Doing sha1 for 3s on 256 size blocks: 324498 sha1's in 2.98s Doing sha1 for 3s on 1024 size blocks: 140371 sha1's in 2.98s Doing sha1 for 3s on 8192 size blocks: 22332 sha1's in 2.98s [/snip] Those numbers are too close to be coincidence. Results with the old Pentium are similar (but much smaller numbers :)) G