Subject: RSA RC5 Key Challege & CPU power
To: Dave Huang <khym@bga.com>
From: Erik E. Fair <fair@clock.org>
List: port-mac68k
Date: 07/14/1997 17:34:28
(please permit me a small digression from the usual topic of this list)

	From:	Dave Huang <khym@bga.com>

	> P.S.	For those of you who have PowerPC machines also, I commend
	> http://rc5.distributed.net/ to your attention. It's one in a series of uses
	> that your machine's idle CPU cycles can be put to. There are others
	> described in http://www.mersenne.org/ (I think this is really cool stuff -
	> part of what the Internet was built for in the first place).

	Awww, even a little 68k machine helps :) I've been running the rc5 thing
	on my 25MHz '040 :) And if you use netbsd-rc5@flame.org as the email
	address, if you find the correct key, the $1000 prize will go to the
	NetBSD Foundation.

Running the Bovine RC5 key space sweeper has been a real eye opener for me,
in terms of exposing the relative CPU power of the various computers I have.

This program is CPU bound - while it is computing, it does almost no I/O, or
system calls. It is heavily dependent (performance-wise) on "rotate"
instructions, which the Intel x86, Motorola 68K & PPC have, and which the Sun
SPARC doesn't have.

Here at the house of too may computers, I've got:

PowerMacintosh 7600/150		MacOS 7.6.1	373 kilokeys/sec
	(dedicated; doing nothing else)

PowerMacintosh 8500/150		MacOS 7.6.1	330 kilokeys/sec
	(my desktop - I'm typing on this)

PowerBook Duo 2300/100		MacOS 7.6.1	220 kilokeys/sec
	(housemate's computer)

SPARCstation 2 (40 MHz SPARC)	SunOS 4.1.4	 30 kilokeys/sec
	(the last SunOS machine)

SPARC LX (50 MHz microSPARC)	NetBSD-current	 29 kilokeys/sec
	(main production)

SPARC LX (50 MHz microSPARC)	NetBSD-current	 29 kilokeys/sec
	(NFS & AppleShare server)

SPARC CLSC (50 MHz microSPARC)	NetBSD-current	 29 kilokeys/sec
	(NetBSD/sparc build system)

Quadra 950 (33 MHz 68040)	MacOS 7.6.1	  9 kilokeys/sec
	(AppleShare server)

Quadra 700 (25 MHz 68040)	MacOS 7.6.1	  6 kilokeys/sec
	(console for all SPARCs)

Sun 3/60 (20 MHz 68020)		NetBSD-current	  2.5 kilokeys/sec
	(NetBSD/sun3 build system)

After looking at this, I decided not to run the Bovine client on the 68K
machines - the Sun 3/60 would turn in one key block every 27 hours or so (and
probably not even that often, since it spends a large percentage of its power
doing the daily NetBSD builds); the PPC machines, by contrast, turn in a key
block every 15 to 20 *minutes*. Actually, the Quadra 950 is just barely worth
it, so maybe I'll run it there.

Look at it this way: one Duo 2300 equals 22 Quadra 950's. One 150 MHz PowerPC
604 equals 37 Quadra 950's. Is it any surprise that I go for recruiting the
PowerPC owners?

The thing is, while my old machines are generally adequate for my usual
computing needs, I find that the price for cycles is a lot better for the new
PowerPC machines than just about anything else, and as soon as NetBSD runs
stably on the PowerMacintosh, I'll dump the SPARCs. They're horridly expensive
per CPU cycle (even on the used market), the Sbus cards are similarly
overpriced compared to what you find in the PCI market. The main thing I do
like about the SPARCs I have is their "lunchbox" form factor, which is easy
to put on a shelf or under a chair; they're quite compact (on the other hand,
the 7[356]00 case is a superior mechanical design - easy to get in & out of).

The 68K machines are probably not even worth the electricity to power them,
if I could consolidate NFS & AppleShare onto a NetBSD system (NetAtalk is
pretty good, but it keeps no "file number" state, and that makes aliases break,
which is unacceptable. The performance also sucks for opening large directories
in the finder, and for "Find File" searches), I'd probably dump them too.
Well, that and getting cross-compilation reliable. Imagine doing daily builds
on a 200 MHz PPC 604e?

The overall point is that Apple & its clone vendors are delivering *incredible*
bang for the buck right now in the PowerPC systems (and from what I hear,
it'll only get better in the fall), and if we could get NetBSD on 'em...

	Erik E. Fair	fair@clock.org