[rc5] Faster clients, etc...why deschall won.

Adam Haberlach HaberlaA at testlab.orst.edu
Sun Jul 6 20:54:28 EDT 1997


- -----Original Message-----
From:	Michael J. Rogan [SMTP:mrogan at fpelectronics.com]
Sent:	Friday, July 04, 1997 8:45 AM
To:	rc5 at llamas.net
Subject:	Re: [rc5] Re: Faster clients

>>This group really needs to invest in faster clients.
>> The only reason the Deschall group was able to stay ahead
>> of the 'foreign' competitors was due to highly optimized

>Deschall was able to draw on a very large group of programmers
>contributing continuous improvements this can't be done without

	Source was only given to people who had signed an agreement/NDA
with Rocke stating that they would not export his code nor share
it with anyone who had not signed the NDA.  I don't think that
there were more then 20 people who had access to the code, and
several of us weren't programmers.
	This was mainly because Rocke didn't want to go to jail for
exporting crypto, but it also kept our fears of a spam-attack
down, since even the protocol was hidden to most people.  At one
time we considered printing up the code in an OCR font and
sending it to someone in South Africa, but we never got around
to it.

	The main reason that Deschall won: we cracked a lot of keys
really fast.  We were able to do that by selecting people who
could produce good cracking code for different systems.  Rocke
produced code that did something like 400 instructions in 233
cycles on the Pentium.  Someone else managed to bit-slice two
different 32-bit clients at the same time on a 64-bit
UltraSparc.  It was these clever hacks that let us do it.
	The other reason was focus on simplicity.  We did not gripe
(much) about stats.  All we wanted were the basics, although
eventually we ended up with much more.  We didn't waste time
with proxies or multi-tiered keyserver systems.  We never had
more then one keyserver, a 486 that kept up with demand due to
Rocke's good design of protocol--it barely loaded the system at
all, even with several thousand computers checking in every
30-50 minutes.  We probably spent more cycles processing stats
then keeping track of keys on the server.  But we spent far MORE
cycles checking keys.

	I think that this effort should work along those lines as well.
Find a few people who really know what they're doing, and let
them have source.  Work on streamlining things.  Don't scale the
servers up.  Scale the protocol down.  Stats are secondary. If
this must be a penis contest, we're trying to produce a flood. 
We don't care about range.  We care about volume.

- ---
Adam Haberlach
Free Lottery Tickets!  Odds: 1/72,000,000,000,000

Version: 2.6.2

To unsubscribe, send email to majordomo at llamas.net with 'unsubscribe rc5' in the body.

More information about the rc5 mailing list