Conference noted::hackers_v1

Great Idea, but can it charge against cost centres?

Also, back in the days of old, I used an Atlas computer (the first paged
machine!) under the Titan operating system. If you wanted to run programs
interactively, you had to log in with /MODE=EXPENSIVE. Your charge centre
was debited accordingly.

Hugh

Of course, RSTS/E did it first -- the development system had
a coin slot (borrowed from a pin-ball machine?) with the label
"Insert 25 cents for 15 small buffers."

But seriously folks, something quite similar was proposed in the
'60's for allocating stand-alone time:  every researcher was
given some fixed "credit" at the beginning of the month, and
could bid on machine time.  Late-night time was cheap, so you
could get a lot of it, while prime-time was expensive, so you
got much less for your budget.

Actually, instead of coin slots and sysgen parameters, you need
agressive system managers who can provide sufficient computing
resources for you to get your work done in reasonable fashion.
but that's another story.

Martin.

Some IBM 370 plug-compatible (maybe an Amdahl) had a "speed" switch on the
front.  You could have it at the normal or the fast setting.  At the fast
setting, the clock ran 50% or so faster.  The catch:  The machines were
leased, and the amount of time you used the machine in each of the two modes
went into your charge at the end of the month.

The theory was that you ran at normal speed most of the time, and switched to
fast mode at the end of the quarter, or whenever you had a lot to do.
						-- Jerry

Speaking of negative time...

	When I was in college our group found a bug in TOPS-20 batch system. 
The batch system would ignore user's diskquota limit. Not only it ignored it, 
but it screwed up user's disk space allocation altogether. We used to run 
recursive batch jobs that created huge log files in user's directory.
After deleting this file the user had plenty of space to work (needless to say 
that at college disk space was one of the most expensive resources). After 
this procedure was followed command @INFORMATION DISK showed negative disk 
pages used. We reported the bug to our computer center, so no one was 
expelled...

					- Victor "innocent" Muslin -

	More time and space hacks:

	On a Xerox Sigma-6 at college a friend of mine wrote a FORTRAN callable
	subroutine that set up a service routine for the "almost out of CPU
	time" interrupt.  The service routine reset the users CPU timer to
	give him another 10 minutes of CPU time...  this repeated every 10
	minutes of CPU.  The routine was called SUPERFLY because the theme
	song of the movie "Superfly" had the words: "time's running out and
	there's no happiness..."

	On the same machine, I discovered that the BASIC subsystem (a command
	line interpreter that looked like Dartmouth BASIC) was immune to
	running out of disk space.  When you created a large enough file
	and went back to the operating system, you had a negative available
	disk space.  By logging off and logging back in you reset your available
	disk space to zero.  Delete the file and your available space goes
	up by the size of the file.  As long as you fill all of the space
	before you log out, you could keep it!

	An interesting property of CP-V (the Xerox operating system) was that
	disk space went to the account that deleted a file even if the file
	was in another account.  When you create files, however, the space
	comes from the account the file is created in!  You could borrow space 
	from another account by writing into it and deleting the file.

			-John A. Wasser (R.I.T. class of 1980)

Well, VAX-Ada is on its way and IT HAS the system for you!   Time-Slicing by
AST's.    Because we run a multiple tasks in a single VMS process, we use 
AST delivery to control the basically round-robin preemptive scheduler.  But
all those AST's going off all the time boost your priority and guarantee you'll
never get outswapped!

All this, and the language of the future too - buy one now and get one ????

/Bevin

re .0 : people who need the priority buy it, while people who don't sell it...

This poses an interesting problem : those who work have to pay to get it done,
while those who just sit around logged on get paid for not wanting a high 
priority.

			Barbara

re: .-1

The offset to the "people who get paid for just sitting around and not
paying for priority" is that they are accountable for their share of the
output/product/service in order to continue to be paid.  Their rational
behavior is to demand exactly enough resources to produce the output
required.  Another name for this concept is efficiency.

Besides the joking around about this, my point is that there is "no free
lunch" when it comes to this business of priorities and access to computer
resources. It's a zero sum game with all incentives matched by disincentives.

A lot of hate that's directed at computers is a reflection not of the
computer or technology-based decisions but of local politics masked by a lot
of obscure computer-babel.

Pat Sweeney

Way back in 1965, the CTSS system at MIT was altered to allow users
to swap prime-time CPU cycles, off-time CPU cycles, disk-space, 
connect time, and any other resource that the system enforced quotas
for,  AT ANY RATE OF EXCHANGE AGREEABLE TO BOTH PARTIES!  Thus it
"made a market" of futures in CPU, ports, disks etc.


Someone familiar with both computer usage, and some techniques for
analyzing the commodities market, wrote an "arbitrage" program that
tried to buy low and sell high.  He left it running in background
permanently and went to New Hampshire for a week.

When he got back, he owned 99% of all the resources on the system for
the next 2 months.  (Did this really happen?  Gee, I heard it more
than once at the time, but it could have been mythical).

Regardless of whether it happened or not, is it fair?

Comments welcome.

Re .9:

Not always so, Pat!

If someone is playing a CPU intensive game, and I wan't to do a few Bliss
compiles, I'll stick them in the dregs queue, and get done faster.  They'll
get done slower, and it doesn't matter.

Applying external value systems onto the work being done changes things,
and so does the existance of unused cycles on a CPU.  Perhaps timesharing
services shouldn't charge for running instead of the null job!?
				/AHM

An interesting example of the difficulties that complex charging policies
can run into:

Back in my hacking days, Princeton had a 360/91.  You could get time on the
thing, or you could use one of a number of free batch services.  For example,
you could run a small WATFOR (FORTRAN) job for free; that service ran - at a
fairly high priority, in fact - every 20 minutes or so.  Among the services
was one whose name now escapes me; call it X.  X allowed you to duplicate a
deck of cards, making all sorts of changes to it - move fields around, sub-
stitute characters, etc. - or to get a listing of a card deck.  X had been
installed for political reasons:  There had been a small stand-alone machine
that had for years run a simpler version of X, and users demanded access to
such a service; the computer center, however, wanted to get rid of the old
machine, which was costing them tons in maintenance.  Hence, service X had
three interesting properties:

	It was free, and had no limits on cards or lines of output;

	Jobs submitted to it ran immediately;

	Jobs submitted to it ran at super-high priority - above all
	other user level stuff.

Now, every other way of punching cards or getting stuff printed cost money,
had limits, and had priorities based on how much you wanted done.

Enter the hack.  The 360/91 ran OS/360 plus LASP.  LASP was an interesting
system:  The ASP was "Attached Processor Support".  ASP was originally a
system that ran on a small "attached processor", and essentially acted
like an OS/360 operator.  You see, OS was really terrible about doing
things like scheduling jobs, running printers, and so on.  ASP took over
all the serial devices, maintained some queues, used operator commands
to control what OS was doing, and grabbed all output destined for serial
devices, queued it, and "did the right thing".  LASP was a software-only
version of ASP:  You booted OS, then ran LASP, which took over, patched
OS so that when it tried to talk to the operator it really called LASP
back, and also took away all OS's serial devices.  Instead, it gave OS
a fake device akin to a mailbox over which "cards" and "line printer
output" flowed.

Now, programs could talk to LASP if they knew how.  Basically, all you
had to do was send properly formatted messages to the "operator's con-
sole" - which was a non-privileged operation.  Among the things you
could get LASP to do for you was take a file and submit it as a job.

Now the picture should be clear:  I want to print a file.  I build it,
leading off with a couple of lines that call up service X and send the
rest of the file to the printer.  I pass the file off to LASP to execute.
Result:  I get my output before anyone else, and I don't even get charged
for it!

Something seems wrong with the economics here, and it's easy to see what
it is:  Why do I charge people for use of line printers and cards?  Pre-
sumably, it's because paper and cards and maintenance on machines cost
money.  But they cost the same amount whether I use X to produce output,
or some other program.  Clearly, the charging policies are irrational.  In
fact, this is exactly a "government subsidy" for certain kinds of "goods" -
and the distortion it introduces in the "marketplace" is inevitable.

						-- Jerry

	When I was in NY last, someone was selling COPB's on a corner
in central park.  Victor, was that you?  If so, mine broke and I want
my $10 back.

	KC