"Toy" computers http://www.conmicro.cx/hercules

From: Raymond Moyers <rmoyers_at_nop.org>
Date: Tue Apr 30 00:08:15 2002

On Monday 29 April 2002 15:35, you wrote:

>> todays mainframes are exteamly muscular

> Indeed. I would probably buy one regardless of the cost that the
> power bill would cause me. :)

 Well the newer ones are getting better, the new fridge sized
 boxes require a fraction of the cooling the old ones did

>> but your statement that the entire campus full of hardware
>> would be emulated on a single peesee, seemed to say
>> differently.

 Ok perspective is needed, here is a picture


 Add in the other stuff that would fill up a room or two on most the other
 buildings on campus, depending on year you have card punches
 3174 term controllers really messy wire closets and racks of IBM MAU's
 ( the token ring equ of a HUB ) this is long before things started
 getting small, a IBM 3864 modem was a hefty 30 pounds or so.

 So what i said was correct, the herc emu is emulating the majority
 of all that stuff, dasd,, tapes,, card deck with an emulated
 "reader" the 3174's and the terminals, that campus full of stuff
 running virtualized on your PC.

> > Remember when the mickysoft press release parroting nattering
> > nabob computer press was declaring the death of the mainframe ?
> They're not still doing it? That surprises me.

 Hard when they are busy posting stories about IBM's double digit
 growth i guess ;=)

( with IBM giving much of the credit to the iron penguins ;=)

> > and very humorous events like when the idiot press would read a
> > product release about NT being ported to an FSIOP card, and run
> > to print "NT Ported to the mainframe ! "
> I would like to see that, actually -- the article, I mean.

 It was this crackpot on infoworld, but he wasnt alone, it was
 at the time when the press would talk about NT everywhere when
 it was knowwhere, i guess doing their part to create the illusion
 of "everyone else, why not you?"

> > A FSIOP card, File Server I/O Processor,

> I suppose such a product would be good if you need it. It could do
> better than to run NT. Maybe they should "port CP/M" to the mainframe.

 Before that OS/2 and Netware was already on them.

> > PC's are certainly as fast per CPU in a general sense
> > as a mainframe, but without the I/O capacity could never
> > hope to replace a mainframe anytime soon.

> I don't know exact numbers, but honestly, the CPU in a modern peesee
> isn't the weak spot at all. Generally there's some kind of bottleneck
> (or five) that needs repaired in the design.

Well a decent modern board is no slouch there either really

Give that a look at what comming w the hammer line

 You cant really trash PC I/O anymore, its right up there
 with everything else and surpasses all the old workstations,
 current PC I/O dont compete with a current mainframe,
 but what ever did ?

 There has been really large R&D sums spent on pushing PC
 performance. and its starting to show bigtime.

 In a way, comparing a mainframe to a PC is like a coal train to a
 dodge viper, sure the viper is faster, but lets see how well it
 does hooked to 80 hopper cars full of coal eh ?

> I wouldn't doubt that it might compare for certain (probably single-user)
> non-io-bound applications. I'm not sure I can make that conclusion for
> a supercomputer at all. When is "yesterday" in this context? :)

 Well litteraly yesterday, dreamworks, pixar etc is going all linux
 for their stuff and the supercomputers people talk about these
 days on the www.top500.org list are linux clusters.

 Another example, look at www.ltsp.org, they are netbooting old
 retired PC's used a diskless xterminals and hanging up to 200
 of them off a modern machine that the apps run on.
 It saves them bucks, and a machine not bloated down
 to a 486 with winblows can hump a good load these days.

> > Compare with the size of bsd/linux/unix that will still run on
> > a machine with 4megs or compare with a mainframe nucleus

> Your comment about mainframes having "stayed small" is oddly amusing,

 Stay small they did, sure the linux kernel tree has grown with the
 in-tree driver count and branches for all the different hardware,
 the same tree builds for a sun alpha or s390.

 but the resultant built kernel has not grown much over the years,
 same for the mainframe, sure it has lots of services hanging around
 it but its core also has stayed very trim.

 One important object of kernel development is for the code to
 get smaller and faster consistent with the other goals.
 Linux, the BSD's and IBM's top line operating systems have
 done well here.

 My firewall DNS mail www and other sundres is still running on an
 old 486 EISA machine, and its just as happy running the same
 kernel and userspace as the 1000mhz linux box with the
 nvidia 3D card, runs fast on slow machines, runs all the faster
 on fast machines.

 The agenda for mickysoft dumbf**kware n co, however, seems
 to be ""cover up any hardware performance gains ( and existing
 hardware) with bloat, forcing the market to buy new boxes
 and end up with no gain at all.""

> the baggage in the design, still hanging around from the beginning.
> (probably not too well-thought-out back then ;) They probably should have
> done something ground-up by now to take advantage of newer cores, bus
> technology, etc.

 My main beaf with x86 is register starvation, at least AMD is doing
 something about it.

 c++ for example, eats a register for "this" and on a register
 starved cpu it hurts far more than others that have more
 registers, this type of thing is perhaps why some fare well
 by compare despite slower clock speeds.

 Its hard to ignore raw speed however, what the x86 lacks
 archtectually its seems to be making it up with brute force.

Received on Tue Apr 30 2002 - 00:08:15 BST

This archive was generated by hypermail 2.3.0 : Fri Oct 10 2014 - 23:34:34 BST