Newsgroups: alt.folklore.computers
Path: nntp.gmd.de!newsserver.jvnc.net!howland.reston.ans.net!usc!
nic-nac.CSU.net!newshub.sdsu.edu!darkstar.UCSC.EDU!news.hal.COM!decwrl!
netcomsv!netcom.com!petrich
From: petr...@netcom.com (Loren Petrich)
Subject: How Do the Old Mainframes Compare to Today's Micros?
Message-ID: <petrichD0sCsq.A11@netcom.com>
Organization: NETCOM On-line Communication Services (408 261-4700 guest)
Date: Wed, 14 Dec 1994 05:34:02 GMT
Lines: 54
I'm asking here because I'm sure that there might be _someone_
around here who would know.
I've been looking in some local libraries for performance stats on
various "classic" mainframes and minicomputers (Eniac, orig. Univac, IBM
360, IBM 370, DEC-10, PDP-10, PDP-11, VAX, Control Data CDC's, etc.),
without much success. Does anyone have any figures or does anyone know of
some books which contain such figures? I ask because I wish to compare
today's microprocessor-based computers to those old models. Do we truly
have a "mainframe on a desktop" nowadays?
Here's some of what I have in mind:
CPU speed (how many operations/s)
RAM size (how many bytes)
Disk-drive capacity (just typical figures)
Comparative bulk and power consumption would also be helpful.
It would be interesting to consider how much electricity a typical
computer chip would consume if it used vacuum tubes (a.k.a. firebottles or
glassfets). 1 million vacuum tubes at 1 watt/tube (probably an
underestimate) adds up to quite a lot. At 10 cents/kwh (a common price),
run 8 hours a day on weekdays, that adds up to a monthly electric bill of
$16,000. And that's not even counting the RAM. 8 megabytes is 64 megabits,
and at 2-3 tubes/bit, this is 100-200 million tubes -- an electric bill of
$2-$3 million a month!
And the bulk? Assume 1 foot by 2 in by 2 in (30 cm by 5 cm by 5
cm) per tube -- one needs to be able to replace bad tubes. The result, 750
m^3, or if fitting on one floor, 20m*20m*2m (60ft*60ft*6ft) -- huge! Its
minimum dimensions would be (9m)^3 ((27)ft^3) -- which means that a ray of
light would take 30 nanoseconds to cross it. This means that the
processing speed in operations/s has to be _less_ than 30 MHz, or else it
could not communicate with itself. And the most recent generation of CPU
chips are _faster_ than that.
And how long could it run before a tube burned out? If it takes, on
the average, 10,000 hours for each one to do so, then the Mean Time
Between Failures would be, for 1 million tubes, 36 seconds! And that
would not even count the time needed to track down the burnt-out tube!
I once saw an amusing parallel that suggested that if cars had
been improved as much as computers, then [well, I forgot, but something
very dramatic].
--
Loren Petrich, the Master Blaster
petr...@netcom.com Happiness is a fast Macintosh
l...@s1.gov And a fast train
Newsgroups: alt.folklore.computers
Path: nntp.gmd.de!newsserver.jvnc.net!howland.reston.ans.net!
news.moneng.mei.com!uwm.edu!lll-winken.llnl.gov!enews.sgi.com!decwrl!
netcomsv!netcom.com!creed
From: cr...@netcom.com (Chris Reed)
Subject: Re: How Do the Old Mainframes Compare to Today's Micros?
Message-ID: <creedD0sD94.6qE@netcom.com>
Organization: is the bane of entropy
References: <petrichD0sCsq.A11@netcom.com>
Date: Wed, 14 Dec 1994 05:43:51 GMT
Lines: 18
In article <petrichD0sCsq....@netcom.com>,
Loren Petrich <petr...@netcom.com> wrote:
> I once saw an amusing parallel that suggested that if cars had
>been improved as much as computers, then [well, I forgot, but something
>very dramatic].
>
How about:
...then a Rolls would cost $100. and would go 2000 miles/hour and would
be 2 inches long.
The Main frame study is more interesting.
--
________________________________________________________________________
|\ \ |C| / /| "The challenge of computer e-mail cr...@netcom.com
| \ \|R|/ / | science is -- How NOT to Phone I'm in the book
| \ |E| / | make a mess of it." Snail Creed Software
|\ \ |E| / /| ...Edsger W. Dijkstra Santa Clara, Ca.
| \ \|D|/ / | 95051-6809
------------------------------------------------------------------------
Path: nntp.gmd.de!newsserver.jvnc.net!howland.reston.ans.net!agate!kahogan
From: kaho...@ieee.EECS.Berkeley.EDU (Kevin A. Hogan)
Newsgroups: alt.folklore.computers
Subject: Re: How Do the Old Mainframes Compare to Today's Micros?
Date: 14 Dec 1994 07:32:13 GMT
Organization: University of California, Berkeley
Lines: 45
Message-ID: <3cm71t$3t8@agate.berkeley.edu>
References: <petrichD0sCsq.A11@netcom.com>
NNTP-Posting-Host: ieee.eecs.berkeley.edu
In article <petrichD0sCsq....@netcom.com>,
Loren Petrich <petr...@netcom.com> wrote:
[ . . . ]
>
> I've been looking in some local libraries for performance stats on
>various "classic" mainframes and minicomputers (Eniac, orig. Univac, IBM
>360, IBM 370, DEC-10, PDP-10, PDP-11, VAX, Control Data CDC's, etc.),
>without much success. Does anyone have any figures or does anyone know of
>some books which contain such figures? I ask because I wish to compare
>today's microprocessor-based computers to those old models. Do we truly
>have a "mainframe on a desktop" nowadays?
One of the books that I used for class this semester, _Computer
Organization & Design: The Hardware/Software Interface_ (D. Patterson
and J. Hennessy, Morgan Kaufmann, 1994), has a small table with
information like what you're looking for on page 40:
UNIVAC I IBM System/360 Model 50
Year 1951 1964
Size (cubic ft.) 1,000 60
Power (watts) 124,500 10,000
Performance (adds/sec) 1,900 500,000
Memory (KB) 48 64
Price $1,000,000 $1,000,000
Adjusted Price (1991 $) $4,533,607 $3,756,502
Price/Performance
vs. UNIVAC 1 318
. . . they go on to list a PDP-8, a Cray-1, the original IBM PC, and a
HP 9000/755. Needless to say, the HP comes out significantly ahead of
anything else, scoring 16,122,356 in the "Price/Performance" column.
Patterson & Hennessy give their sources for the information in this
chart as the Computer Museum in Boston and the Producer Price Index for
Industrial Commodities.
-- Kevin Hogan
--
------------------------------------------------------------------------------
Kevin Hogan -- kaho...@ucsee.EECS.Berkeley.EDU -- (510) 664-2533
Vice President, University of California Society of Electrical Engineers
WWW Home Page: http://ucsee.EECS.Berkeley.EDU/~kahogan/index.html
Path: nntp.gmd.de!newsserver.jvnc.net!nntpserver.pppl.gov!princeton!udel!
news.sprintlink.net!pipex!swrinde!news.uh.edu!news.uth.tmc.edu!
admin5.hsc.uth.tmc.edu!not-for-mail
From: jmayn...@admin5.hsc.uth.tmc.edu (Jay Maynard)
Newsgroups: alt.folklore.computers
Subject: Re: How Do the Old Mainframes Compare to Today's Micros?
Date: 14 Dec 1994 05:01:06 -0600
Organization: University of Texas - Houston, General Administration
Lines: 33
Message-ID: <3cmj9i$nv4@admin5.hsc.uth.tmc.edu>
References: <petrichD0sCsq.A11@netcom.com> <creedD0sD94.6qE@netcom.com>
NNTP-Posting-Host: admin5.hsc.uth.tmc.edu
In article <creedD0sD94....@netcom.com>, Chris Reed <cr...@netcom.com> wrote:
>In article <petrichD0sCsq....@netcom.com>,
>Loren Petrich <petr...@netcom.com> wrote:
>> I once saw an amusing parallel that suggested that if cars had
>>been improved as much as computers, then [well, I forgot, but something
>>very dramatic].
>How about:
>...then a Rolls would cost $100. and would go 2000 miles/hour and would
>be 2 inches long.
...and blow up once a year, killing all its occupants.
>The Main frame study is more interesting.
Well, as long as you understand you're comparing apples and oranges in terms
of capability...
The first machine I did as an MVS systems type was a 370/158AP. The base
machine was what defined the MIPS, both in the IBM and VAX worlds (the
11/780 was defined as one VAX MIPS because, it was claimed, it was about as
powerful as a /158). It had 6 MB RAM (four of which was an Intel add-on
box), and 4 8-volume strings of 3350 disk drives. Each volume was about
300 MB (my memory is failing me, and I don't recall how many cylinders a
3350 had; each cylinder was 30 tracks, of 19069 bytes max each). The machine
had 6 channels, 5 of which were block channels that could run at 1.5
MB/second. The AP designated an attached processor, which was a second /158
CPU that shared main memory but had no I/O functions of its own. (The 3081,
circa 1979, was the first IBM symmetric multiprocessor; earlier MPs didn't
share I/O.) The machine was built in the mid-70s; we got it in 1981.
--
Jay Maynard, EMT-P, K5ZC, PP-ASEL | Never ascribe to malice that which can
jmayn...@admin5.hsc.uth.tmc.edu | adequately be explained by stupidity.
8 November 1994: It's the guns, stupid! (Thanks, America...)
Path: nntp.gmd.de!newsserver.jvnc.net!nntpserver.pppl.gov!princeton!udel!
news.sprintlink.net!howland.reston.ans.net!cs.utexas.edu!uunet!
news.uiowa.edu!news
From: jo...@pyrite.cs.uiowa.edu (Douglas W. Jones,201H MLH,3193350740,3193382879)
Newsgroups: alt.folklore.computers
Subject: Re: How Do the Old Mainframes Compare to Today's Micros?
Date: 14 Dec 1994 16:01:59 GMT
Organization: University of Iowa, Iowa City, IA, USA
Lines: 55
Distribution: world
Message-ID: <3cn4tn$dr2@nexus.uiowa.edu>
References: <petrichD0sCsq.A11@netcom.com>
NNTP-Posting-Host: pyrite.cs.uiowa.edu
From article <petrichD0sCsq....@netcom.com>,
by petr...@netcom.com (Loren Petrich):
>
> I'm asking here because I'm sure that there might be _someone_
> around here who would know.
>
> I've been looking in some local libraries for performance stats on
> various "classic" mainframes and minicomputers (Eniac, orig. Univac, IBM
> 360, IBM 370, DEC-10, PDP-10, PDP-11, VAX, Control Data CDC's, etc.),
The CDC 6600, fastest computer on earth between its introduction in the
mid 1960's and the advent of newer supercomputers in the '70s was able
to sustain about 1 megaflop.
I've benchmarked my laptop at the same speed. My laptop uses a SPARC II
processor; it's about 5 times faster than my old IBM RT workstation
and a lot slower than the IBM RS6000 that serves as our departmental
hub.
A CDC 6600 could easily handle 50 to 500 time-shared users (depending on
what the users were doing), but my laptop has never supported more than
two or three timeshared users -- one at the keyboard and one or two
connected by ethernet.
> Do we truly have a "mainframe on a desktop" nowadays?
Yup. In terms of compute power, there's no question about it. In terms
of disk space, there's no question about it.
> CPU speed (how many operations/s)
1 megaflop used to be the definition of a supercomputer
now it's commonplace on your desktop.
> RAM size (how many bytes)
1 megabyte used to be considered large for a mainframe
now it's commonplace on your desktop.
> Disk-drive capacity (just typical figures)
32 megabytes is a typical figure for the storage capacity of
an old multi-platter, 14 inch diameter disk pack. A typical
minicomputer from 1973 would have two such drives. A typical
mainframe would have had a "disk farm" in a computer center
with from 4 to 16 of these. Thats 1/2 gigabyte. Today, my
laptop has close to half that, and you'll find plenty of desktop
machines with more.
>
> Comparative bulk and power consumption would also be helpful.
A small minicomputer from 1965 ran at 700 watts. A large
minicomputer from 1973 ran at 2000 watts. Large mainframes
drew quite a bit more than that. My laptop runs at 55 watts,
at full display brightness, full CPU speed, and charging a
battery at the same time.
Doug Jones
jo...@cs.uiowa.edu
Newsgroups: alt.folklore.computers
Path: nntp.gmd.de!Germany.EU.net!howland.reston.ans.net!spool.mu.edu!uwm.edu!
lll-winken.llnl.gov!enews.sgi.com!decwrl!netcomsv!netcom.com!richgr
From: ric...@netcom.com (Rich Greenberg)
Subject: Re: How Do the Old Mainframes Compare to Today's Micros?
Message-ID: <richgrD0tGoM.6wu@netcom.com>
Organization: Sorry, left it in my other pants.
References: <petrichD0sCsq.A11@netcom.com> <creedD0sD94.6qE@netcom.com>
<3cmj9i$nv4@admin5.hsc.uth.tmc.edu>
Date: Wed, 14 Dec 1994 19:55:34 GMT
Lines: 22
In article <3cmj9i$...@admin5.hsc.uth.tmc.edu> jmayn...@admin5.hsc.uth.tmc.edu
(Jay Maynard) writes:
[...]
>MB/second. The AP designated an attached processor, which was a second /158
>CPU that shared main memory but had no I/O functions of its own. (The 3081,
>circa 1979, was the first IBM symmetric multiprocessor; earlier MPs didn't
>share I/O.) The machine was built in the mid-70s; we got it in 1981.
No, the 3081 was far from the first. The 360/65, 67, 370/158, 168 were
all available as MPs, with the second processor having full i/o
capability. They were essentially 2 complete machines joined by extra
circutry which allowed each CPU to shoulder tap the other, and to get
to each others storage.
The 3081 was the first dyadic machine, where the two processors were
closer together than the earlier MPs and unlike the earlier MPs could
not be run as 2 seperate machines by flipping a few switches.
--
Rich Greenberg Work: TBA. Know anybody needing a VM guru?
N6LRT TinselTown, USA Play: ric...@netcom.com 310-649-0238
Pacific time. I speak for myself & my dogs only. Canines: Chinook & Husky(RIP)
Newsgroups: alt.folklore.computers
Path: nntp.gmd.de!newsserver.jvnc.net!nntpserver.pppl.gov!princeton!udel!
news.intercon.com!panix!zip.eecs.umich.edu!newsxfer.itd.umich.edu!
europa.eng.gtefsd.com!howland.reston.ans.net!spool.mu.edu!uwm.edu!
lll-winken.llnl.gov!enews.sgi.com!decwrl!netcomsv!netcom.com!richgr
From: ric...@netcom.com (Rich Greenberg)
Subject: Re: How Do the Old Mainframes Compare to Today's Micros?
Message-ID: <richgrD0tGvE.77J@netcom.com>
Organization: Sorry, left it in my other pants.
References: <petrichD0sCsq.A11@netcom.com> <3cn4tn$dr2@nexus.uiowa.edu>
Date: Wed, 14 Dec 1994 19:59:38 GMT
Lines: 18
In article <3cn4tn$...@nexus.uiowa.edu> jo...@pyrite.cs.uiowa.edu
(Douglas W. Jones,201H MLH,3193350740,3193382879) writes:
>From article <petrichD0sCsq....@netcom.com>,
[...]
>> Do we truly have a "mainframe on a desktop" nowadays?
>
>Yup. In terms of compute power, there's no question about it. In terms
>of disk space, there's no question about it.
True enough on those two items. The one aspect of a mainframe that the
desktop units have not come close to matching (as yet) is i/o bandwidth.
A large IBM (or IBM compatible) can have 512 channels, each of which
can be passing data at 10 mbytes/sec, all at the same time.
--
Rich Greenberg Work: TBA. Know anybody needing a VM guru?
N6LRT TinselTown, USA Play: ric...@netcom.com 310-649-0238
Pacific time. I speak for myself & my dogs only. Canines: Chinook & Husky(RIP)
Path: nntp.gmd.de!Germany.EU.net!EU.net!howland.reston.ans.net!swrinde!
news.uh.edu!news.uth.tmc.edu!admin5.hsc.uth.tmc.edu!not-for-mail
From: jmayn...@admin5.hsc.uth.tmc.edu (Jay Maynard)
Newsgroups: alt.folklore.computers
Subject: Re: How Do the Old Mainframes Compare to Today's Micros?
Date: 14 Dec 1994 18:54:34 -0600
Organization: University of Texas - Houston, General Administration
Lines: 20
Message-ID: <3co44a$4r7@admin5.hsc.uth.tmc.edu>
References: <petrichD0sCsq.A11@netcom.com> <creedD0sD94.6qE@netcom.com>
<3cmj9i$nv4@admin5.hsc.uth.tmc.edu> <richgrD0tGoM.6wu@netcom.com>
NNTP-Posting-Host: admin5.hsc.uth.tmc.edu
In article <richgrD0tGoM....@netcom.com>,
Rich Greenberg <ric...@netcom.com> wrote:
>In article <3cmj9i$...@admin5.hsc.uth.tmc.edu> jmayn...@admin5.hsc.uth.tmc.edu
(Jay Maynard) writes:
>>(The 3081,
>>circa 1979, was the first IBM symmetric multiprocessor; earlier MPs didn't
^^^^^^^^^
>>share I/O.)
>No, the 3081 was far from the first. The 360/65, 67, 370/158, 168 were
>all available as MPs, with the second processor having full i/o
>capability. They were essentially 2 complete machines joined by extra
>circutry which allowed each CPU to shoulder tap the other, and to get
>to each others storage.
True...but they were not symmetric in that the I/O channels were not shared.
Yes, they could be switched, but you could not have both processors doing
I/O to the same channel set at the same time.
--
Jay Maynard, EMT-P, K5ZC, PP-ASEL | Never ascribe to malice that which can
jmayn...@admin5.hsc.uth.tmc.edu | adequately be explained by stupidity.
8 November 1994: It's the guns, stupid! (Thanks, America...)
Newsgroups: alt.folklore.computers
Path: nntp.gmd.de!Germany.EU.net!EU.net!sun4nl!cwi.nl!dik
From: d...@cwi.nl (Dik T. Winter)
Subject: Re: How Do the Old Mainframes Compare to Today's Micros?
Message-ID: <D0twF9.8F7@cwi.nl>
Sender: n...@cwi.nl (The Daily Dross)
Nntp-Posting-Host: boring.cwi.nl
Organization: CWI, Amsterdam
References: <petrichD0sCsq.A11@netcom.com> <3cn4tn$dr2@nexus.uiowa.edu>
Date: Thu, 15 Dec 1994 01:35:32 GMT
Lines: 45
In article <3cn4tn$...@nexus.uiowa.edu> jo...@pyrite.cs.uiowa.edu
(Douglas W. Jones,201H MLH,3193350740,3193382879) writes:
> The CDC 6600, fastest computer on earth between its introduction in the
> mid 1960's and the advent of newer supercomputers in the '70s was able
> to sustain about 1 megaflop.
I think it was able to sustain more, but you had to go through hoops to
get more. For Cybers (the true ones, i.e. not those with a unified
functional unit) could get very close to 1/(cycle time) flops. And I
think that the 6600 had something like 100 nset major cycle, giving
10 megaflop. My off-the-cuff figures where always: realistic about
one third of that for tuned assembly and one sixth for compiled code.
...
> A CDC 6600 could easily handle 50 to 500 time-shared users (depending on
> what the users were doing), but my laptop has never supported more than
> two or three timeshared users -- one at the keyboard and one or two
> connected by ethernet.
From some old discussions I understood that with SunOS you would have
problems with even 16 simultaneous processes.
...
> > Do we truly have a "mainframe on a desktop" nowadays?
>
> Yup. In terms of compute power, there's no question about it. In terms
> of disk space, there's no question about it.
But in terms of the number of simultaneous users the answer is no.
...
> > RAM size (how many bytes)
> 1 megabyte used to be considered large for a mainframe
> now it's commonplace on your desktop.
Indeed. Early Cybers where maxed at 131072 60-bit words.
...
> > Comparative bulk and power consumption would also be helpful.
> A small minicomputer from 1965 ran at 700 watts. A large
> minicomputer from 1973 ran at 2000 watts. Large mainframes
> drew quite a bit more than that. My laptop runs at 55 watts,
> at full display brightness, full CPU speed, and charging a
> battery at the same time.
>
But, for those old machines, display brightness was just a drop in
the ocean of course.
--
dik t. winter, cwi, kruislaan 413, 1098 sj amsterdam, nederland, +31205924098
home: bovenover 215, 1025 jn amsterdam, nederland; e-mail: d...@cwi.nl
Newsgroups: alt.folklore.computers
Path: nntp.gmd.de!Germany.EU.net!EU.net!howland.reston.ans.net!
vixen.cso.uiuc.edu!uwm.edu!lll-winken.llnl.gov!enews.sgi.com!decwrl!
netcomsv!netcom.com!petrich
From: petr...@netcom.com (Loren Petrich)
Subject: Re: How Do the Old Mainframes Compare to Today's Micros?
Message-ID: <petrichD0u0yM.E9x@netcom.com>
Organization: NETCOM On-line Communication Services (408 261-4700 guest)
References: <petrichD0sCsq.A11@netcom.com> <3cm71t$3t8@agate.berkeley.edu>
Date: Thu, 15 Dec 1994 03:13:34 GMT
Lines: 46
In article <3cm71t$...@agate.berkeley.edu>,
Kevin A. Hogan <kaho...@ieee.EECS.Berkeley.EDU> wrote:
>In article <petrichD0sCsq....@netcom.com>,
>Loren Petrich <petr...@netcom.com> wrote:
[Present-day personal computers vs. classic mainframes, minis...]
> One of the books that I used for class this semester, _Computer
>Organization & Design: The Hardware/Software Interface_ (D. Patterson
>and J. Hennessy, Morgan Kaufmann, 1994), has a small table with
>information like what you're looking for on page 40:
> UNIVAC I IBM System/360 Model 50
> Year 1951 1964
> Size (cubic ft.) 1,000 60
> Power (watts) 124,500 10,000
> Performance (adds/sec) 1,900 500,000
> Memory (KB) 48 64
> Price $1,000,000 $1,000,000
> Adjusted Price (1991 $) $4,533,607 $3,756,502
> Price/Performance
> vs. UNIVAC 1 318
Thanx to you and the others who dug up this nice info. For
comparison, here are some specs of my Quadra 650:
Year: 1993
Size [CPU only]: 2 ft^3
Power: I'm not sure, but for the CPU, is probably not greater than 100
watts (it's cooler than my monitor).
Memory (KB): 8192
Performance: probably a few million adds/s (its 68040 processor is
rated at 33 MHz).
Price [CPU only]: $3000 (1993 dollars)
The processing speed/price ratio for Quadra/Sys360 is about
10,000, while for Quadra/Univac, it is about 1,000,000. For memory, both
turn out to be 100,000 [I'm assuming equal fractions of the cost go to
each feature].
--
Loren Petrich, the Master Blaster
petr...@netcom.com Happiness is a fast Macintosh
l...@s1.gov And a fast train
Newsgroups: alt.folklore.computers
Path: nntp.gmd.de!Germany.EU.net!EU.net!howland.reston.ans.net!
vixen.cso.uiuc.edu!uwm.edu!lll-winken.llnl.gov!enews.sgi.com!decwrl!netcomsv!
netcom.com!petrich
From: petr...@netcom.com (Loren Petrich)
Subject: Re: How Do the Old Mainframes Compare to Today's Micros?
Message-ID: <petrichD0u1FF.FK1@netcom.com>
Organization: NETCOM On-line Communication Services (408 261-4700 guest)
References: <petrichD0sCsq.A11@netcom.com> <3cn4tn$dr2@nexus.uiowa.edu>
<richgrD0tGvE.77J@netcom.com>
Date: Thu, 15 Dec 1994 03:23:39 GMT
Lines: 36
In article <richgrD0tGvE....@netcom.com>,
Rich Greenberg <ric...@netcom.com> wrote:
>In article <3cn4tn$...@nexus.uiowa.edu> jo...@pyrite.cs.uiowa.edu
(Douglas W. Jones,201H MLH,3193350740,3193382879) writes:
>>From article <petrichD0sCsq....@netcom.com>,
>>> Do we truly have a "mainframe on a desktop" nowadays?
>>Yup. In terms of compute power, there's no question about it. In terms
>>of disk space, there's no question about it.
>True enough on those two items. The one aspect of a mainframe that the
>desktop units have not come close to matching (as yet) is i/o bandwidth.
>A large IBM (or IBM compatible) can have 512 channels, each of which
>can be passing data at 10 mbytes/sec, all at the same time.
That implies a total throughput of 5 gigabytes/s -- one wonders
what kind of bus architecture can sustain that kind of throughput, and
how it manages to fit with the rest of the system.
I do recall that IBM mainframes (System 360 and its successors)
have a sort of DMA [Direct Memory Access] I/O: the CPU just issues "Start
I/O", "Halt I/O", and "Test I/O" instructions, with appropriate Channel
Command Words, and the peripheral I/O processors do the rest. So that may
be one of the reasons that IBM mainframes can maintain that kind of
performance.
This sort of I/O capacity suggests that big mainframes may have a
continued existence as giant servers, if nothing else. Is that what
mainframes tend to be nowadays?
--
Loren Petrich, the Master Blaster
petr...@netcom.com Happiness is a fast Macintosh
l...@s1.gov And a fast train
Path: nntp.gmd.de!newsserver.jvnc.net!news.cac.psu.edu!news.pop.psu.edu!
psuvax1!uwm.edu!spool.mu.edu!bloom-beacon.mit.edu!uhog.mit.edu!
europa.eng.gtefsd.com!hookup!olivea!uunet!news.uiowa.edu!icaen!dsiebert
From: dsieb...@icaen.uiowa.edu (Doug Siebert)
Newsgroups: alt.folklore.computers
Subject: Re: How Do the Old Mainframes Compare to Today's Micros?
Date: 15 Dec 1994 17:36:03 GMT
Organization: Iowa Computer Aided Engineering Network, University of Iowa
Lines: 32
Message-ID: <3cpuq3$t9d@news.icaen.uiowa.edu>
References: <petrichD0sCsq.A11@netcom.com> <3cn4tn$dr2@nexus.uiowa.edu>
<D0twF9.8F7@cwi.nl>
NNTP-Posting-Host: l_ecn02.icaen.uiowa.edu
d...@cwi.nl (Dik T. Winter) writes:
>...
> > A CDC 6600 could easily handle 50 to 500 time-shared users (depending on
> > what the users were doing), but my laptop has never supported more than
> > two or three timeshared users -- one at the keyboard and one or two
> > connected by ethernet.
>From some old discussions I understood that with SunOS you would have
>problems with even 16 simultaneous processes.
>...
> > > Do we truly have a "mainframe on a desktop" nowadays?
> >
> > Yup. In terms of compute power, there's no question about it. In terms
> > of disk space, there's no question about it.
>But in terms of the number of simultaneous users the answer is no.
Depends on what you are doing. Those old-time users weren't running emacs,
C++ compilers, etc. For something that relates better to what was done back
then as far as CPU requirements you see increases. The ISCA BBS supports well
over 1000 simultaneous users on an HP 9000/710 (50MHz PA-RISC) HP doesn't
make workstations that slow anymore, with a newer machine and more memory it
could top 2000 users. What is done there is probably not that far off from
what old mainframe users did in terms of the types of workload the machine is
being asked to do.
--
Doug Siebert | I have a proof that everything I have stated above
dsieb...@isca.uiowa.edu | is true, but this .sig is too small to contain it.
Path: nntp.gmd.de!newsserver.jvnc.net!howland.reston.ans.net!cs.utexas.edu!
uunet!news.uiowa.edu!news
From: jo...@pyrite.cs.uiowa.edu (Douglas W. Jones,201H MLH,3193350740,3193382879)
Newsgroups: alt.folklore.computers
Subject: Re: How Do the Old Mainframes Compare to Today's Micros?
Date: 15 Dec 1994 19:50:26 GMT
Organization: University of Iowa, Iowa City, IA, USA
Lines: 47
Distribution: world
Message-ID: <3cq6m2$pa9@nexus.uiowa.edu>
References: <3cpuq3$t9d@news.icaen.uiowa.edu>
NNTP-Posting-Host: pyrite.cs.uiowa.edu
From article <3cpuq3$...@news.icaen.uiowa.edu>,
by dsieb...@icaen.uiowa.edu (Doug Siebert):
>
> Depends on what you are doing. Those old-time users weren't running emacs,
> C++ compilers, etc.
You bet! A 50 user CDC 6600 was had most of the users running ICE
(the In Core Editor), to cut down on swapping. Interactive programming
was as rare on that beast as it was on an IBM 360, so what you did was
write your FORTRAN program, then compile it, and if it compiled
correctly, run it. At any time, only a few users were actually running
programs.
And then, there were the marvelous kluges done to make the 6600 do
timesharing at all. Remember, that machine had no paging, just a
base-bound register pair. As a result, ICE sat in core to serve all
interactive editing users, and another two or three core partitions were
set aside for user programs. Fast context switching (initiated by one
of the peripheral processors) was used to switch back and forth between
the current jobs in the different partitions, but if a job overstayed its
welcome in core, a PP generated a core dump (as card images on a disk
file, of all things) and submitted it at the end of a batch job queue,
along with appropriate, automatically generated control cards to cause
it to be loaded and restarted where it left off.
To make a CDC 6600 handle 500 users, as they did at the Illinois Plato
Project, you had to write your own special system, deny users the right
to run FORTRAN jobs, and limit them to an average of about 2000
instructions per second (2 TIPS) per user. I wrote an interactive
graphic game on Plato that managed to stay under 5 TIPS, HiVolts, a
possible precursor of PacMan, but that was quite an exercise. Most
interactive games ran at closer to 20 TIPS, and some required considerably
more.
> For something that relates better to what was done back
> then as far as CPU requirements you see increases. The ISCA BBS
> supports well over 1000 simultaneous users on an HP 9000/710
> (50MHz PA-RISC).
And I'll guarantee you that you'd have difficulty offering that level of
service on a CDC 6600 without a custom crafted runtime system. ISCA BBS
uses plain vanilla UNIX under their user interface. Nothing so nice was
available as a foundation on the old mainframes (ok, I'll make an exception
for MULTICS and admit that Amdahl did a nice port of UNIX to their 370
clones).
Doug Jones
jo...@cs.uiowa.edu
Newsgroups: alt.folklore.computers
Path: nntp.gmd.de!newsserver.jvnc.net!netnews.upenn.edu!msunews!uwm.edu!
vixen.cso.uiuc.edu!howland.reston.ans.net!gatech!bloom-beacon.mit.edu!
news.kei.com!world!decwrl!netcomsv!netcom.com!petrich
From: petr...@netcom.com (Loren Petrich)
Subject: Re: How Do the Old Mainframes Compare to Today's Micros?
Message-ID: <petrichD0yozo.GGB@netcom.com>
Organization: NETCOM On-line Communication Services (408 261-4700 guest)
References: <3cpuq3$t9d@news.icaen.uiowa.edu> <3cq6m2$pa9@nexus.uiowa.edu>
Date: Sat, 17 Dec 1994 15:43:00 GMT
Lines: 58
In article <3cq6m2$...@nexus.uiowa.edu>,
Douglas W. Jones,201H MLH,3193350740,3193382879 <jo...@pyrite.cs.uiowa.edu>
wrote:
>From article <3cpuq3$...@news.icaen.uiowa.edu>,
>by dsieb...@icaen.uiowa.edu (Doug Siebert):
>> Depends on what you are doing. Those old-time users weren't running emacs,
>> C++ compilers, etc.
>You bet! A 50 user CDC 6600 was had most of the users running ICE
>(the In Core Editor), to cut down on swapping.
And I presume it was some sort of line-mode (pure-CLI) editor,
like UNIX ed/ex, instead of a semi-GUI editor like emacs (a GUI editor
for character-based displays).
And I presume that its OS was pure CLI, and a clunky one at that.
... Interactive programming
>was as rare on that beast as it was on an IBM 360, so what you did was
>write your FORTRAN program, then compile it, and if it compiled
>correctly, run it. At any time, only a few users were actually running
>programs.
I presume from this discussion that each user could have only one
process at a time -- that one would edit, quit from the edit, compile,
and run -- all in sequence (I've gone through that myself with other
systems, like an IBM mainframe running VM/CMS and a VAX running VMS,
though on the latter, I could edit on one terminal, and compile and run
on another).
>And then, there were the marvelous kluges done to make the 6600 do
>timesharing at all. Remember, that machine had no paging, just a
>base-bound register pair. [details deleted]
And I wonder what kind of response it had to its users.
>To make a CDC 6600 handle 500 users, as they did at the Illinois Plato
>Project, you had to write your own special system, deny users the right
>to run FORTRAN jobs, and limit them to an average of about 2000
>instructions per second (2 TIPS) per user. I wrote an interactive
>graphic game on Plato that managed to stay under 5 TIPS, HiVolts, a
>possible precursor of PacMan, but that was quite an exercise. Most
>interactive games ran at closer to 20 TIPS, and some required considerably
>more.
I guess not many people were running DOOM on that system :-)
Not to mention that DOOM requires a graphics-display terminal.
DOOM is, of course, a game that requires a _lot_ of calculation
to generate its graphics -- and generate them in real time (it reportedly
requires at least an Intel 486).
--
Loren Petrich, the Master Blaster
petr...@netcom.com Happiness is a fast Macintosh
l...@s1.gov And a fast train