On Sat, Feb 11, 2006 at 08:06:39PM -0800, Jason Thorpe wrote: > This needs to be done. rnd_add_uint32() seems to do an awful lot of > work. It already defers the worst of the work of actually mixing in the samples, but it would be feasible to defer the delta calculations too. The estimator is pretty much known to be bogus anyway, and represents the majority of the work presently done here. I wouldn't miss it if a gig-E just assumed one bit of entropy per sample, or even per sample batch. Rui, as a first part of the testing, please measure with entropy estimation on and off (rndctl -e/-E). That will guide any next steps. -- Dan.
Attachment:
pgpzPw4m06D2O.pgp
Description: PGP signature