Safe decryption with GnuPG?

[David Shaw]

On Thu, Feb 14, 2008 at 08:44:32AM -0600, Robert J. Hansen wrote:
>> Full disk encryption makes the system unnecessarily slow,
>> especially if applied to swap space.
>
> Not necessarily so.  A lot of people make a big deal out of a couple of 
> papers published on how much whole-disk encryption slows down OpenBSD, but 
> the flip side to that is the file and network systems of OpenBSD are not as 
> efficient as those of many other OSes.  If you've done your own empirical 
> tests with your own OS and discovered it's too slow, then by all means, 
> it's too slow.  Otherwise, you may wish to do some empirical tests.

Definitely.

It is also important to test with different methods of disk
encryption.  On most systems there are a number of tunable parameters
that can significantly affect performance.  For example, AES is going
to be faster than 3DES.  Does the kernel do disk readahead?  If so,
how much?  Another question is how you're using it: a small file,
written once and read many times will be much faster on average than a
large file written once and read back once.  It's not as simple as
just saying "unnecessarily slow".

I worked on a disk encryption system a while back.  In design
discussions, we spent a lot of time discussing potential performance
issues and how that would affect the users of the system.  Finally, I
advised this:

1) Understand how you're going to use it.  What are you protecting
   against?  Given that encrypted disks are mounted and readable on a
   running system (effectively bypassing the encryption), often you're
   really just protecting against theft or a disk "sprouting legs" and
   going missing.

2) Tune the system for best performance for your particular usage.

3) Measure that performance.

4) Quantify in terms of money just how much the performance hit hurts
   you.  Call this "A".

5) Find out how much it would cost to hire someone big with a gun to
   stand in front of the server 24/7.  Call this "B".

6) Which is cheaper?

I've found that at the end of step 6, people tend to ask the smart
questions, and really start to understand step 1.

David