On Fri, 9 Sep 2011, Damien Miller wrote: > Hi, > > This diff needs testing, particularly on systems that support hardware > acceleration of AES via the OpenSSL EVP layer (e.g. Intel Core i7). > It uses OpenSSL's EVP AES API rather than the lower-level one and should > give an opportunity for the acceleration to work. > > A useful benchmark would be > > dd if=/dev/arandom bs=100000 count=1000 | time ssh localhost "cat > /dev/null" > > before and after (you will need passwordless authentication setup so > as not to stall at the password prompt)
Thanks to everyone that tested. The redux is that going to EVP doesn't seem to provide any discernable performance benefit on capable CPUs while imposing a performance penalty on everyone else. I don't intend to pursue this further. -d
