Re: [TLS] TLS, PKI,

[Martin Rex <[e-mail]>]

Steingruebl, Andy wrote:
> 
> Actually, I think that the recent work that the Google guys did
> - http://www.imperialviolet.org/2010/06/25/overclocking-ssl.html
> - might have some impact on this.  There is some cost to using TLS.   

Of course there is some cost to using TLS, but computationally it
has been marginal for common web uses for many years.  Where you
notice it is with downloads of large amounts of data (several MBytes)
over high-speed links (100 MBit/s+), because the raw encryption
speed of the symmetric ciphers becomes visible.

The performance of TLS with software-crypto was perfectly acceptable
6 years ago on x86 and x86_64 bit Hardware platforms.  I 
All performance problem I've seen so far were the result of bugs
and software design flaws, but not problems of the technology or
lack of speed of the underlying hardware (maybe with the exception
of a few RISC-based unix boxen for which a decent crypto speed with
portable C code has been notoriously difficult to achieve).

The most difficult issue with larger servers is not the crypto,
but correct and efficient TLS session cache manangement.

And there still is a lot of (non-browser) software out there where
TLS session caching doesn't work, is defective in certain
configurations (Microsoft IIS) or is not implemented (apache mod_proxy).


-Martin