[RC5] DES-Test, Day 2
gwar at silcom.com
Fri Jan 8 14:34:01 EST 1999
>Date: Fri, 8 Jan 1999 14:35:20 -0500 (EST)
>From: John Campbell <jcampbel at lynn.ci-n.com>
>Subject: Re: [RC5] DES-Test, Day 2
>On Thu, 7 Jan 1999, Jim C. Nasby wrote:
>> time. We've completed 19.32% of t he keyspace, sent 131.03% (we've begun
> We've recycled more of the keyspace than we've actually cracked!?
>There's going to be some *serious* duplication of work going on here,
>because I suspect that the vast majority of those recycled keys are just
>waiting in buffers to be cracked or flushed, and not actually lost blocks.
> It seems to me that this means that we need smaller blocks and
>smaller buffers to ensure that the block with the key in it gets sent back
>ASAP. These contests are getting short enough that the total duration of the
>contest is approaching the time it takes for a slow machine to crack a
>single block. So... if we want those slow machines to continue to be useful,
>we need finer granularity, with smaller blocks available, and practically no
>buffering on the slower machines.
> As things are, I'm going to be running most of my machines with a
>buffer size of 1, because that'll give them at least a chance of getting a
>block or two in before the end of the contest, and maybe even before the
>block they're working on gets sent to someone else.
Yup. It just makes sense. If you're in a contest that's scheduled to end
in a time-frame measured in hours it's not very intelligent to buffer more
blocks than you could check in a few hours. Less if the keyservers can
handle the load and you've got a 24/7 connection. I've got one machine
that's rarely connected which I loaded up with a thousand RC5 blocks just
so I wouldn't have to worry about it running out. Whenever it's connected
I do a flush. I figure I'll get thru those thousand blocks long before the
RC5-64 contest is over. When it comes to DES, tho, that would be stupid.
It would take one of my slower machines ~10 days to get thru 1000 2^28
sized DES blocks. Even my fastest machine would take ~2 days by which time
the contest would (hopefully) be over.
Also, since the block buffer is a FILO structure, a computer set to buffer
1000 incoming blocks and 10 outgoing blocks will have 990 blocks that are
never touched (unless it can't connect to flush its 10 blocks).
Just put the client in Modem Dialup Lurk mode and make sure the dialer
kicks in and disconnects automatically (in a short amount of time). Figure
out how many blocks you can crack in an hour or so and set the buffers to
hold that many blocks. My slowest machine, a 486DX4/100 (yeah, it's slow
but it makes a decent fax server for the office), takes 38 minutes to get
thru a 2^28 size block of DES keys so, when I get around to optimizing my
clients, it'll get a buffer setting of 2:2. The Cyrix by my desk gets a
block done in about 17 minutes so I'll set its buffer to 4:4. My K6-2's
crank out about 4 times that so they'll probably get set to 15:15. Etc.,
gwar at silcom.com
"Three monkeys, ten minutes."
To unsubscribe, send 'unsubscribe rc5' to majordomo at lists.distributed.net
rc5-digest subscribers replace rc5 with rc5-digest
More information about the rc5