use poly1305-AES for faster MACs
seems to be way faster than hmac-shaXXX. how fast is it compared to GMAC (see aes-gcm)?
https://en.wikipedia.org/wiki/Poly1305-AES + http://cr.yp.to/mac/53.html
Maybe "comparable" in the absence of hardware support. Bernstein gives comprehensive benchmarks. The best case shown below works out at 4.5 cycles/byte, v.s. 6.7 for GMAC.
http://www.cryptopp.com/benchmarks-p4.html http://cr.yp.to/mac/speed-fireball.txt.bz2
However full aes-gcm i.e. including encryption, with aes-ni acceleration, is said to achieve 3.5 cycles/byte or lower. (Disclaimer: on something a bit more recent than a P4).
Maybe setup time is better for poly1305 (1000 vs 9000 cycles?), but for 64K chunks I think that's also "comparable".