[Cryptography] cheap sources of entropy

Theodore Ts'o tytso at mit.edu
Sun Feb 2 15:38:53 EST 2014

On Mon, Feb 03, 2014 at 04:08:08AM +1000, James A. Donald wrote:
> OK then,  In the case that the real machine is CPU bound, and not IO
> bound or cache thrashing, then you will not see turbulence
> randomness at the VM level.
> However, when starting up a new program, you generally are IO bound,
> though by no means guaranteed to be IO bound.

*Your* VM may be I/O bound, but other guest OS's, running on other
VM's, may be CPU bound at the same time.  In fact, if the hosting
service is being smart, they might try to start, and possibly migrate,
VM's which are mostly CPU bound to be on the same physical machine as
other VM's which are mostly I/O bound, so they can more efficiently
use 100% of all of the host OS's resources.  In the ideal world, you
want each physical machine to be using close to 100% of all available
memory, CPU, disk time, and networking bandwidth.  Companies that can
do this will can afford to sell VM's cheaper than their competition,
and guess what, will tend to be the ones that survive and propser.

And therefore, you may find that even though you are I/O bound, your
disk interrupts in your guest OS are being delivered to you subject to
certain quantization effects.  And even if it isn't doing so today,
you might find that tomorrow, your hosting service may change how they
do things to more efficiently use their physical servers...

    	  	     	     	- Ted

More information about the cryptography mailing list