Re: HTTP/2 and Pervasive Monitoring

On 20 August 2014 10:32, Mark Nottingham <mnot@mnot.net> wrote:

> I'm curious; do Ilari's numbers <
> http://www.w3.org/mid/20140817120844.GA1346@LK-Perkele-VII> change your
> mind at all?
>

For many servers, the cost of applying the actual cipher is not the main
impost of doing TLS.   The main impost for java servers is just bringing
the data into user memory in the first place so that it can be sliced,
passed to a TLS encoder and then at the very least copied if not encoded.

If we don't have to apply a cipher at all then we can do direct writes with
memory mapped files that move data from file system to network without
transiting user memory.  To apply the null cipher, there will be a data
copy into user space, a data copy from input to output buffers and a data
copy out of user memory.   These 3 copies are almost certainly a lot more
expensive than the cipher (which is done during one of the copies).

I'll try to find the time to see if I can get some hard numbers for what is
the impact of applying even a null cipher in java.    But I'm dubious if
the cost of terminating TLS will be insignificant any time soon.

regards







-- 
Greg Wilkins <gregw@intalio.com>
http://eclipse.org/jetty HTTP, SPDY, Websocket server and client that scales
http://www.webtide.com  advice and support for jetty and cometd.