k-e-i-t-h-m-o at e-x-m-s-f-t.com
Mon Mar 29 04:16:27 PST 2004
Here's an interesting technique for extracting entropy from air
turbulence in disk drives: http://world.std.com/~dtd/random/forward.pdf
or http://world.std.com/~dtd/random/forward.ps. Linux's /dev/random uses
disk timings as one source of entropy for its pool, so it indirectly
uses this technique already.
Of course, as Robert said, this doesn't help the diskless systems.
Imagine a headless, diskless server used as a compute node in a cluster.
The only source of entropy would be from the network. This would be an
unacceptably lame source.
Another problem with diskless systems: They have no place to save a
random seed between reboots. This means they always boot into exactly
the same machine state. (Cluster administrators tend to think this is a
great "feature", and it probably is in many ways.)
With the recent growth of relatively inexpensive clusters, I think this
must be a solved problem. A bit of searching on Google was less than
encouraging. The general consensus seems to be "some systems have
practically zero entropy -- use a hardware RNG".
More information about the hlfs-dev