[email protected] (Timothy Sipples) writes:
> 2. Average costs versus marginal costs. If you simply take the total
> expense and divide that up into chargebacks, you've got a problem. That'll
> cause very bad behavior as users try to flee what they see as high costs
> (average costs) which really truly aren't (they're actually marginal
> costs). One better way (albeit not perfect) is to charge an unavoidable,
> universal "membership fee" (per employee, for example) plus a variable
> rate, with the variable rate equal to true marginal costs. That's similar
> to your electric bill -- a "connection fee" plus a charge per kilowatt
> hour. Bonus points for peak and non-peak pricing. Non-peak could even be
> zero.

way back when IBM leased everything ... leasing rates was month
for 1st 40hr@week, then lower for 41-80hrs, and then lower for
81-120hr, and then lowest for 121-168hrs.

IBM leasing was based on cpu meter which ran whenever the processor
and/or channels were busy.

in the 60s, early days of the virtual machine online commercial service
bureaus (CP67, precursor to vm370) tried to encourage offshift uses with
lower rate ... however, initially the useage didn't cover keeping the
service available 7x24 ... but w/o having the service available 7x24, it
was hard to encourage off-shift uses. one of the issues was to minimize
all off-shift, non-essential expenses ... aka darkroom operation,
eliminate requirements for operator, automatic reboot/restart in case of
fault, etc.

One of the issues was normal channel programs that would be able to
accept terminal connection and incoming characters would also keep the
meter running. So another hack was terminal controller channel command
word that would allow channel to go idle (letting the leasing cpu meter
come to stop) ... but would come active whenever there was incoming
characters (system was 7x24 available, but meter wouldn't be running
unless something was actually going on).

A consequence then of the 7x24 operation was that they had no window for
hardware maintenance. As a result by the early 70s, those operations had
enhanced online service so that it supported loosely-coupled operation
with non-disruptive migration of processes between systems (so systems
could be taken offline for hardware maintenace, w/o no disruption for
online users).

One of the features of the cpu meter (for leasing charges) was that both
cpu and all channels had to be idle for at least 400mills before the
meter actually stopped. An indication that the MVS SRM had been designed
during the leasing period was that it had a timer interval that wokeup
every 400mills ... whether or not anything was going on (aka if MVS was
running ... the cpu meter was guarenteed to always run regardless of
whether anything was going on or not).

Total cost has been taken to extreme in modern mega-datacenters ... with
hundreds of thousands or even millions of processors. cost of machines
have dropped so low that electricity and cooling have started to become
dominate costs. mega-datacenters are being designed to minimize cooling
costs and located in areas that can best take advantage of natural
cooling and also have lowest (& most dependable) electrical costs.

-- 
virtualization experience starting Jan1968, online at home since Mar1970

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to