/LMAR=0/XLINE=3/FONT#0=BASL30/FONT#1=BASI30/FONT#2=BASB30/FONT#10=BAXM30/FONT#11=ZERO30/FONT#5=MICR25/FONT#6=NGB30



␈↓ ↓H␈↓¬␈↓ α≡OFFICE MEMORANDUM  *  STANFORD UNIVERSITY  *  OFFICE MEMORANDUM

␈↓ ↓H␈↓ε␈↓ β=STANFORD LOW OVERHEAD TIME-SHARING SYSTEM

␈↓ ↓H␈↓α␈↓ ¬VFebruary 16, 1979




␈↓ ↓H␈↓αTo:␈↓ αH␈↓Professor G. Lieberman

␈↓ ↓H␈↓␈↓αFrom:␈↓ αH␈↓Ralph Gorin and John McCarthy

␈↓ ↓H␈↓αSubject:␈↓ αH␈↓Request for more LOTS hardware

␈↓ ↓H␈↓In␈α∩the␈α∩academic␈α∩year␈α∩1977-78,␈α∩the␈α∩LOTS␈α∩computer␈α∩facility␈α∩took␈α∩over␈α∩the␈α∩full␈α∩load␈α∩of␈α∩class
␈↓ ↓H␈↓computing␈α∂(with␈α∂anticipated␈α∂minor␈α∂exceptions)␈α⊂and␈α∂a␈α∂substantial␈α∂load␈α∂of␈α⊂unsponsored␈α∂research
␈↓ ↓H␈↓and␈α∃student␈α∃projects.␈α∃ By␈α∃the␈α∃end␈α∃of␈α∃Winter␈α∃Quarter,␈α∃the␈α∃full␈α∃complement␈α∃of␈α∃authorized
␈↓ ↓H␈↓equipment␈αwas␈αoperating.␈α We␈αbelieve␈αthat␈αthe␈αavailability␈αof␈αinteractive␈αcomputing␈αto␈α
all␈αclasses
␈↓ ↓H␈↓has␈α⊃expanded␈α⊃the␈α∩use␈α⊃of␈α⊃computers␈α∩and␈α⊃increased␈α⊃its␈α∩usefulness␈α⊃for␈α⊃learning.␈α∩ However,␈α⊃the
␈↓ ↓H␈↓computer␈αis␈α
now␈αoverloaded,␈αand␈α
we␈αhave␈α
good␈αevidence␈αthat␈α
additional␈αcomputer␈αequipment␈α
will
␈↓ ↓H␈↓substantially␈αincrease␈αthe␈αquality␈αof␈αservice␈α
we␈αcan␈αgive.␈α Software␈αimprovements␈αcan␈αalso␈α
provide
␈↓ ↓H␈↓better service, and some are being implemented.

␈↓ ↓H␈↓We request authority to make the following acquisitions:

␈↓ ↓H␈↓        1.␈α⊂Upgrade␈α⊃the␈α⊂processor␈α⊂from␈α⊃a␈α⊂2050␈α⊂to␈α⊃a␈α⊂2060.␈α⊂ This␈α⊃involves␈α⊂replacing␈α⊂most␈α⊃of␈α⊂the
␈↓ ↓H␈↓electronics of the computer and costs $50,000.

␈↓ ↓H␈↓        2.␈α⊃Increase␈α∩of␈α⊃memory␈α∩from␈α⊃512K␈α∩words␈α⊃to␈α⊃1024K␈α∩words.␈α⊃ We␈α∩have␈α⊃a␈α∩proposal␈α⊃from
␈↓ ↓H␈↓Ampex␈αto␈αdo␈αthis␈αfor␈α$115,000,␈αand␈αa␈αproposal␈αfrom␈αDigital␈αEquipment␈αCorporation␈αto␈αdo␈αit␈αfor
␈↓ ↓H␈↓$194,000.␈α Digital␈αis␈αalso␈αpreparing␈αanother␈αplan␈αto␈αdo␈αit␈αfor␈αabout␈α$140,000.␈α Both␈αvendors␈αhave
␈↓ ↓H␈↓o≥ered␈αshort␈αterm␈α
leases␈αby␈αwhich␈αwe␈α
can␈αevaluate␈αthe␈αe≥ectiveness␈α
of␈αthe␈αmemory␈α
in␈αincreasing
␈↓ ↓H␈↓system throughput.

␈↓ ↓H␈↓        3. A second tape drive at $15,000.

␈↓ ↓H␈↓        4. Tax, shipping, insurance and installation charges on the above.

␈↓ ↓H␈↓Total: Approximately $195,000 to $220,000, depending on our choice of vendor.

␈↓ ↓H␈↓In␈α∞addition␈α
to␈α∞the␈α
purchase␈α∞cost,␈α
there␈α∞will␈α
be␈α∞an␈α
additional␈α∞maintenance␈α
cost␈α∞of␈α
approximately
␈↓ ↓H␈↓$10,000 per year.

␈↓ ↓H␈↓Adding␈αto␈αthe␈αmain␈αmemory␈αis␈αthe␈αbest␈αtechnical␈αsolution␈αto␈αour␈αoverloading␈αproblems␈αand␈αit␈αcan
␈↓ ↓H␈↓be␈α
done␈α
for␈α
Fall␈αQuarter.␈α
 This␈α
upgrade␈α
should␈αincrease␈α
computer␈α
throughput␈α
adequately␈αto␈α
serve
␈↓ ↓H␈↓the␈α⊂present␈α⊂level␈α⊂of␈α⊂demand␈α⊂from␈α⊂the␈α∂academic␈α⊂users␈α⊂of␈α⊂LOTS.␈α⊂ In␈α⊂the␈α⊂absence␈α⊂of␈α∂increased
␈↓ ↓H␈↓demands,␈α∀we␈α∀believe␈α∀that␈α∀the␈α∀system␈α∀with␈α∪this␈α∀upgrade␈α∀will␈α∀deliver␈α∀nearly␈α∀twice␈α∀as␈α∪much
␈↓ ↓H␈↓throughput (65 to 70% of all CPU time) under "worst-case" loads as it does presently (35%).



␈↓ ↓H␈↓αProfessor G. Lieberman␈↓ ¬VFebruary 16, 1979␈↓ 
nPage 2␈↓ 


␈↓ ↓H␈↓Making␈α⊂LOTS␈α∂better␈α⊂will␈α∂tempt␈α⊂some␈α∂people␈α⊂to␈α∂make␈α⊂more␈α∂extensive␈α⊂use␈α∂of␈α⊂the␈α⊂facility.␈α∂ To
␈↓ ↓H␈↓restrict␈αthe␈αgrowth␈αof␈α
new␈αuses,␈αwe␈αhave␈α
limits␈αon␈αterminal␈αtime␈α
and␈αthe␈αlimited␈αnumber␈αof␈α
public
␈↓ ↓H␈↓terminals.␈α Restrictions␈αto␈αlimit␈α
the␈αin∨ux␈αof␈αadditional␈αclasswork␈α
may␈αbe␈αnecessary;␈αbut␈αwe␈α
should
␈↓ ↓H␈↓be␈α
careful␈αneither␈α
to␈αpropose␈α
nor␈αtolerate␈α
limitations␈αthat␈α
adversely␈αa≥ect␈α
the␈α
academic␈αprogram.
␈↓ ↓H␈↓At␈αsome␈αtime␈αin␈αthe␈αfuture,␈αthe␈αsingle␈αLOTS␈αmainframe␈αwill␈αnot␈αby␈αitself␈αbe␈αadequate␈αto␈αservice
␈↓ ↓H␈↓all␈α⊂instructional␈α⊂computing;␈α⊂when␈α⊂necessary,␈α⊂the␈α∂LOTS-style␈α⊂of␈α⊂computing␈α⊂can␈α⊂be␈α⊂extended␈α∂to
␈↓ ↓H␈↓further systems.

␈↓ ↓H␈↓The detailed reasons for this request follow.



␈↓ ↓H␈↓αProfessor G. Lieberman␈↓ ¬VFebruary 16, 1979␈↓ 
nPage 3␈↓ 


␈↓ ↓H␈↓BACKGROUND:

␈↓ ↓H␈↓LOTS␈α∞is␈α∂operating␈α∞a␈α∞large␈α∂DECsystem␈α∞2050␈α∞computer␈α∂system␈α∞to␈α∞provide␈α∂interactive␈α∞computing
␈↓ ↓H␈↓service␈α∞to␈α∞support␈α∂instruction␈α∞and␈α∞unsponsored␈α∞research␈α∂at␈α∞Stanford␈α∞University.␈α∞ Of␈α∂the␈α∞11,000
␈↓ ↓H␈↓students␈αat␈αthe␈αUniversity␈αnearly␈α3,000␈αuse␈αLOTS␈αeach␈αquarter.␈α LOTS␈αis␈αused␈αfor␈αinstruction␈αin
␈↓ ↓H␈↓Computer␈α∩Science,␈α∪Engineering,␈α∩Statistics,␈α∩Education,␈α∪and␈α∩the␈α∩social␈α∪sciences.␈α∩ In␈α∪addition␈α∩to
␈↓ ↓H␈↓instruction,␈αLOTS␈αis␈α
also␈αused␈αfor␈α
unsponsored␈αresearch␈αand␈α
to␈αsome␈αextent␈α
for␈αword-processing
␈↓ ↓H␈↓by students writing dissertations and papers.

␈↓ ↓H␈↓The␈α⊂LOTS␈α⊂system␈α⊂hardware␈α⊂presently␈α⊂includes␈α∂a␈α⊂DECsystem-2050␈α⊂processor,␈α⊂512␈α⊂K␈α⊂words␈α∂of
␈↓ ↓H␈↓main␈α∪memory,␈α∩three␈α∪200-megabyte␈α∪disk␈α∩drives␈α∪on␈α∩two␈α∪channels,␈α∪one␈α∩tape␈α∪drive,␈α∪and␈α∩eighty
␈↓ ↓H␈↓terminal ports.

␈↓ ↓H␈↓Two␈α⊂of␈α∂the␈α⊂three␈α∂disk␈α⊂drives␈α∂are␈α⊂in␈α∂use␈α⊂for␈α∂public␈α⊂storage␈α∂and␈α⊂swapping.␈α∂ The␈α⊂third␈α⊂disk␈α∂is
␈↓ ↓H␈↓presently␈αbeing␈αused␈αfor␈αlarge,␈αinfrequently␈αchanged␈αdata␈α≡les,␈αprincipally␈αsystem␈αaccounting␈αdata
␈↓ ↓H␈↓and program sources.

␈↓ ↓H␈↓THE PROBLEM:  Demand exceeds supply.

␈↓ ↓H␈↓Demand␈αfor␈αLOTS␈αchronically␈αsuper-saturates␈αthe␈αe≥ective␈αsupply␈αof␈αservices.␈α We␈αmust␈αimprove
␈↓ ↓H␈↓the␈α
ratio␈αof␈α
e≥ective␈αthroughput␈α
to␈αuser␈α
demands.␈α The␈α
system␈αsaturates␈α
to␈αthe␈α
point␈αwhere␈α
people
␈↓ ↓H␈↓≡nd␈αthat␈αit␈αis␈αno␈αlonger␈αuseful.␈α They␈αwould␈αquit␈αusing␈αLOTS␈αunder␈αthose␈αcircumstances␈αbut␈αfor
␈↓ ↓H␈↓external pressure from due dates.

␈↓ ↓H␈↓The overload at LOTS is apparent in several e≥ects:
␈↓ ↓H␈↓␈↓ αH1.  Lengthy delays for response from the computer,
␈↓ ↓H␈↓␈↓ αH2.  Long queues of students waiting to use terminals,
␈↓ ↓H␈↓␈↓ αH3.  Occasional long delays for printed output,
␈↓ ↓H␈↓␈↓ αH4.  Long waits to access the tape drive, and
␈↓ ↓H␈↓␈↓ αH5.  Shortage of disk space.

␈↓ ↓H␈↓The␈α∞≡rst␈α∂of␈α∞these␈α∞conditions␈α∂is␈α∞the␈α∞most␈α∂distressing.␈α∞ Moreover,␈α∞each␈α∂of␈α∞the␈α∞other␈α∂symptoms␈α∞of
␈↓ ↓H␈↓overload␈α∀can␈α∀be␈α∀traced␈α∃to␈α∀the␈α∀general␈α∀ine≠ciency␈α∃of␈α∀the␈α∀computer␈α∀system␈α∃under␈α∀overload
␈↓ ↓H␈↓conditions.

␈↓ ↓H␈↓A␈αmore␈αe≠cient␈αsystem␈αwill␈αgive␈αbetter␈αresponse,␈αget␈αpeople␈αo≥␈αthe␈αterminals␈αquicker,␈αshorten␈αthe
␈↓ ↓H␈↓queue␈αof␈α
people␈αwaiting␈αfor␈α
terminals,␈αand␈αprovide␈α
better␈αprinter␈αthroughput␈α
on␈αthe␈α
printers␈αwe
␈↓ ↓H␈↓already␈α⊗have.␈α⊗ Less␈α⊗obviously,␈α∃poor␈α⊗system␈α⊗throughput␈α⊗is␈α∃preventing␈α⊗us␈α⊗from␈α⊗taking␈α∃full
␈↓ ↓H␈↓advantage of all the disk space we already have.  (Some further explanation is made below.)

␈↓ ↓H␈↓RESTRICTIONS ON DEMAND

␈↓ ↓H␈↓On␈αthe␈αdemand␈αside,␈αwe␈αhave␈αplaced␈αrestrictions␈αon␈αhow␈αmuch␈αconsole␈αtime␈αeach␈αuser␈αis␈αallowed
␈↓ ↓H␈↓per␈α∂week.␈α⊂ We␈α∂have␈α∂reason␈α⊂to␈α∂believe␈α∂that␈α⊂these␈α∂restrictions␈α∂were␈α⊂not␈α∂entirely␈α⊂e≥ective␈α∂Spring
␈↓ ↓H␈↓quarter.␈α However,␈αa␈αbetter␈αjob␈αcan␈αbe␈αdone,␈αand␈αit␈αwill␈αbe␈αdone␈αthis␈αFall.␈α We␈αare␈αtightening␈αup
␈↓ ↓H␈↓the account-making process this summer, to better prepare for Fall quarter.



␈↓ ↓H␈↓αProfessor G. Lieberman␈↓ ¬VFebruary 16, 1979␈↓ 
nPage 4␈↓ 


␈↓ ↓H␈↓A␈α∞recent␈α∞episode␈α
shows␈α∞how,␈α∞despite␈α∞our␈α
controls,␈α∞we␈α∞have␈α∞failed␈α
to␈α∞reduce␈α∞demand.␈α∞ We␈α
don't
␈↓ ↓H␈↓want␈α
to␈α∞repeat␈α
the␈α∞experience␈α
 we␈α∞ had␈α
 with␈α
the␈α∞Human␈α
Biology␈α∞4␈α
course␈α∞that␈α
put␈α∞200␈α
novice
␈↓ ↓H␈↓users,␈α∂all␈α⊂doing␈α∂resource-consuming␈α⊂plots,␈α∂onto␈α⊂LOTS␈α∂at␈α∂about␈α⊂the␈α∂sixth␈α⊂week␈α∂of␈α⊂the␈α∂quarter.
␈↓ ↓H␈↓However,␈αgiven␈α
that␈αLOTS␈αis␈α
charged␈αwith␈α
supplying␈αall␈αinstructional␈α
computing,␈αwe␈α
have␈αvery
␈↓ ↓H␈↓little␈αmeans␈αto␈αprevent␈αa␈αrecurrence.␈α When␈αProfessor␈αDurham␈α≡rst␈αapproached␈αus␈αin␈αthis␈αmatter,
␈↓ ↓H␈↓it␈αwas␈α
his␈αintention␈α
to␈αuse␈αthe␈α
computer␈αin␈α
early␈αMay,␈αand␈α
this␈αlooked␈α
reasonable␈αto␈αus.␈α
 However,
␈↓ ↓H␈↓he␈α⊃didn't␈α∩≡nish␈α⊃his␈α⊃special␈α∩programs␈α⊃until␈α∩about␈α⊃May␈α⊃15,␈α∩so␈α⊃the␈α⊃students␈α∩in␈α⊃Human␈α∩Bio␈α⊃4
␈↓ ↓H␈↓contributed an extra burden to the usual end-of-quarter rush.

␈↓ ↓H␈↓We␈α
hope␈α
that␈αas␈α
the␈α
faculty␈α
gains␈αa␈α
more␈α
thorough␈α
understanding␈αof␈α
the␈α
capabilities␈α
and␈αlimits␈α
of
␈↓ ↓H␈↓the␈α∩LOTS␈α∪resource␈α∩such␈α∪incidents␈α∩will␈α∪diminish␈α∩in␈α∪frequency␈α∩and␈α∪impact.␈α∩ One␈α∪of␈α∩LOTS'
␈↓ ↓H␈↓functions␈α∞is␈α∞to␈α
provide␈α∞for␈α∞innovative␈α∞pedagogical␈α
use␈α∞of␈α∞the␈α∞computer,␈α
so␈α∞we␈α∞must␈α∞expect␈α
that
␈↓ ↓H␈↓additional␈αcourses␈αwill␈α"discover"␈αus.␈α Furthermore,␈αas␈αmore␈αstudents␈αat␈αStanford␈αbecome␈αaware␈αof
␈↓ ↓H␈↓the␈αvariety␈αof␈αservices␈αthat␈αcan␈αbe␈αobtained␈αat␈αLOTS,␈αwe␈αexpect␈αto␈αsee␈αan␈αincrease␈αin␈αthe␈αservices
␈↓ ↓H␈↓delivered␈α
outside␈α
of␈α
required␈αclasswork.␈α
 Some␈α
other␈α
universities␈αhave␈α
forecast␈α
their␈α
growth␈αrate
␈↓ ↓H␈↓for␈αinstructional␈αand␈αresearch␈α
computing␈αat␈α35%␈αper␈αyear.␈α
 Although␈αthe␈αone␈αmainframe␈αat␈α
LOTS
␈↓ ↓H␈↓will␈αnot␈αbe␈α
adequate␈αinde≡nitely,␈αthe␈αLOTS-style␈α
of␈αcomputing␈αcan␈α
be␈αextended␈αto␈αfurther␈α
systems
␈↓ ↓H␈↓when necessary.

␈↓ ↓H␈↓We␈αpropose␈αto␈αcontinue␈αthe␈αconsole-time␈αallocation␈αsystem␈αthat␈αwe␈αstarted␈αthis␈αspring.␈α There␈αwill
␈↓ ↓H␈↓be some changes to the system, including the following:

␈↓ ↓H␈↓␈↓ αH1. No carry-forward of unused weekly allocations to the next week.

␈↓ ↓H␈↓␈↓ αH2.␈α∩Time␈α∩allocations␈α∩to␈α∩students␈α∩in␈α∩each␈α∩course␈α∩will␈α∩be␈α∩based␈α∩on␈α∩the␈α⊃computational
␈↓ ↓H␈↓␈↓ αHintensity␈α
of␈α
the␈α
particular␈αcourse,␈α
rather␈α
than␈α
on␈α
units␈αalone.␈α
 This␈α
way,␈α
a␈α
course␈αsuch␈α
as
␈↓ ↓H␈↓␈↓ αHHuman␈αBiology␈α4␈αA␈αand␈αB␈α(a␈αtotal␈αof␈α8␈αunits),␈αwouldn't␈αautomatically␈αget␈αan␈αallocation
␈↓ ↓H␈↓␈↓ αHof computer time twice that of CS105.

␈↓ ↓H␈↓␈↓ αH3.␈α∂A␈α∞tightening␈α∂of␈α∞the␈α∂account␈α∞issuing␈α∂process␈α∞to␈α∂eliminate␈α∞questionable␈α∂accounts␈α∞and
␈↓ ↓H␈↓␈↓ αHpoaching.

␈↓ ↓H␈↓␈↓ αH4.␈αEncouragement␈α
of␈αsome␈α
unsponsored␈αresearch␈α
users␈αto␈α
≡nd␈αa␈α
little␈αmoney␈α
and␈αto␈α
avail
␈↓ ↓H␈↓␈↓ αHthemselves of SCIP's services.

␈↓ ↓H␈↓␈↓ αH5. More e≥ective enforcement of the per-session time limits.

␈↓ ↓H␈↓␈↓ αH6.␈α
Establishment␈αof␈α
a␈αterminal␈α
reservation␈αsystem,␈α
in␈αwhich␈α
a␈αterminal␈α
can␈α
be␈αreserved
␈↓ ↓H␈↓␈↓ αHfor␈α∂a␈α∂speci≡c␈α∞time␈α∂in␈α∂the␈α∂future.␈α∞ This␈α∂may␈α∂permit␈α∂people␈α∞to␈α∂budget␈α∂their␈α∂time␈α∞more
␈↓ ↓H␈↓␈↓ αHe≥ectively.

␈↓ ↓H␈↓␈↓ αH7.␈α
Possibly,␈α∞extending␈α
the␈α∞allocation␈α
system␈α
to␈α∞directly␈α
limiting␈α∞line␈α
printer␈α∞output␈α
and
␈↓ ↓H␈↓␈↓ αHcpu time in addition to terminal connect time.



␈↓ ↓H␈↓αProfessor G. Lieberman␈↓ ¬VFebruary 16, 1979␈↓ 
nPage 5␈↓ 


␈↓ ↓H␈↓INCREASING THE SUPPLY:  An e≥ective prescription

␈↓ ↓H␈↓There␈α∂are␈α∂two␈α∞ways␈α∂to␈α∂increase␈α∂the␈α∞supply␈α∂of␈α∂computer␈α∂resources.␈α∞ First,␈α∂we␈α∂can␈α∂employ␈α∞better
␈↓ ↓H␈↓software␈α∪that␈α∩delivers␈α∪more␈α∪of␈α∩the␈α∪hardware␈α∩system's␈α∪capabilities␈α∪to␈α∩the␈α∪users;␈α∪but␈α∩software
␈↓ ↓H␈↓techniques␈α
alone␈α
cannot␈α
produce␈α
the␈α∞drastic␈α
increase␈α
in␈α
e≥ectiveness␈α
that␈α
LOTS␈α∞needs.␈α
 Second,
␈↓ ↓H␈↓we␈α
can␈α
buy␈αadditional␈α
computer␈α
hardware.␈α The␈α
remedy␈α
we␈αpropose␈α
includes␈α
both␈α
software␈αand
␈↓ ↓H␈↓hardware.

␈↓ ↓H␈↓Projects␈α∂to␈α∂build␈α∂a␈α∞terminal␈α∂reservation␈α∂system␈α∂and␈α∞to␈α∂implement␈α∂the␈α∂resource␈α∂limitations␈α∞were
␈↓ ↓H␈↓discussed␈α
above.␈α
 The␈α
following␈α
projects␈α
are␈α
underway␈α
to␈α
increase␈α
the␈α
e≥ectiveness␈α
of␈α
particular
␈↓ ↓H␈↓software areas:

␈↓ ↓H␈↓␈↓ αH1.␈α The␈αdevelopment␈αof␈αa␈αnew␈αimplementation␈αof␈αPASCAL␈αfor␈αuse␈αby␈αthe␈αintroductory
␈↓ ↓H␈↓␈↓ αHand␈α∞intermediate␈α∞Computer␈α∞Science␈α
(and␈α∞related␈α∞Electrical␈α∞Engineering)␈α∞courses.␈α
 This
␈↓ ↓H␈↓␈↓ αHproject␈αis␈αbeing␈αdone␈αin␈αresponse␈αto␈αtwo␈αoperational␈αdi≠culties␈αthat␈αhave␈αbeen␈α
observed
␈↓ ↓H␈↓␈↓ αHwith␈αthe␈αSAIL␈αlanguage␈αimplementation␈αat␈αLOTS.␈α These␈αproblems␈αwith␈αSAIL␈αthat␈αwe
␈↓ ↓H␈↓␈↓ αHhope␈α∪to␈α∪eliminate␈α∪in␈α∪the␈α∪new␈α∪implementation␈α∪of␈α∪PASCAL,␈α∪are:␈α∪ Incomprehensible
␈↓ ↓H␈↓␈↓ αHdiagnostic␈αerror␈αmessages,␈αwhich␈αhinder␈αstudents'␈αprogramming␈αdebugging,␈αand␈αthe␈αuse
␈↓ ↓H␈↓␈↓ αHof␈α∪the␈α∩linkage␈α∪editor␈α∪(LINK)␈α∩which␈α∪is␈α∩a␈α∪large␈α∪and␈α∩ine≠cient␈α∪consumer␈α∪of␈α∩system
␈↓ ↓H␈↓␈↓ αHresources.

␈↓ ↓H␈↓␈↓ αH2.␈α
Changes␈α
to␈α
the␈α
operating␈α
system␈α
to␈α
make␈α
it␈α
allocate␈α
main␈α
memory␈α
and␈α
other␈α
system
␈↓ ↓H␈↓␈↓ αHresources␈α
more␈α
e≥ectively.␈α
 The␈α
LOTS␈α
sta≥␈α
is␈αtoo␈α
small␈α
to␈α
get␈α
into␈α
this␈α
kind␈αof␈α
tinkering
␈↓ ↓H␈↓␈↓ αHtoo␈αdeeply,␈αso␈α
we␈αare␈αin∨uencing␈α
DEC␈αto␈αmake␈α
the␈αperformance␈αimprovements␈α
that␈αwe
␈↓ ↓H␈↓␈↓ αHthink are necessary.

␈↓ ↓H␈↓Most␈αof␈αthe␈αincrease␈αin␈αthroughput␈αwill␈αcome␈αfrom␈αnew␈αhardware,␈αparticularly␈αfrom␈αadding␈αmore
␈↓ ↓H␈↓memory.  The new hardware we seek includes:

␈↓ ↓H␈↓␈↓ αH1. Upgrade to the 2060 central processing unit,

␈↓ ↓H␈↓␈↓ αH2. Addition of 512 K words of memory, to double LOTS' main memory size,

␈↓ ↓H␈↓␈↓ αH3. Addition of one tape drive, and

␈↓ ↓H␈↓␈↓ αH4.␈αNo␈αgrowth,␈αor␈αvery␈αsmall␈αgrowth,␈αin␈αthe␈αnumber␈αof␈αterminals␈αattached␈αto␈αLOTS.␈α No
␈↓ ↓H␈↓␈↓ αHadditional terminals for LOTS are requested at this time.

␈↓ ↓H␈↓The ≡rst step: a new processor

␈↓ ↓H␈↓Additional␈αmemory␈αis␈αour␈αmost␈αpressing␈αneed,␈αbut␈αit␈αis␈αnecessary␈αto␈αupgrade␈αthe␈αcentral␈αprocessor
␈↓ ↓H␈↓to␈αthe␈α2060␈αlevel␈αwhen␈αwe␈αadd␈αmemory.␈α The␈α2060␈αincludes␈αa␈αfaster␈αand␈αmore␈αversatile␈αCPU␈α
that
␈↓ ↓H␈↓can␈α
support␈α
more␈αmemory␈α
than␈α
our␈α2050␈α
has.␈α
Although␈αwe␈α
don't␈α
presently␈αneed␈α
a␈α
faster␈αCPU,␈α
the
␈↓ ↓H␈↓other features of the 2060 are necessary:

␈↓ ↓H␈↓␈↓ αH1.␈α∂LOTS␈α∂needs␈α∂the␈α⊂2060␈α∂to␈α∂keep␈α∂up␈α⊂with␈α∂DEC's␈α∂future␈α∂software␈α⊂developments.␈α∂ We



␈↓ ↓H␈↓αProfessor G. Lieberman␈↓ ¬VFebruary 16, 1979␈↓ 
nPage 6␈↓ 

␈↓ ↓H␈↓␈↓ αHmust␈α_follow␈α→DEC's␈α_operating␈α_system␈α→developments␈α_because␈α_we␈α→have␈α_insu≠cient
␈↓ ↓H␈↓␈↓ αHmanpower to undertake the long-term development and support of such a system.

␈↓ ↓H␈↓␈↓ αH2.␈αBy␈αextending␈α
the␈αvirtual␈αaddress␈αspace,␈α
the␈α2060␈αallows␈αus␈α
to␈αintegrate␈αmore␈αthan␈α
two
␈↓ ↓H␈↓␈↓ αHdisk␈α∂drives␈α∂into␈α∂the␈α∂publicly␈α∂accessible␈α∂disk␈α∂(one␈α∂of␈α∂our␈α∂disks␈α∂is␈α⊂presently␈α∂considered
␈↓ ↓H␈↓␈↓ αHprivate because of this restriction).

␈↓ ↓H␈↓␈↓ αH3.␈αThe␈αextended␈αvirtual␈αaddress␈αspace␈αof␈α
the␈α2060␈αallows␈αroom␈αfor␈αthe␈αdata␈α
tables␈αthat
␈↓ ↓H␈↓␈↓ αHare␈α⊂needed␈α⊂to␈α⊂describe␈α⊂the␈α∂memory␈α⊂that␈α⊂we␈α⊂want␈α⊂to␈α∂add␈α⊂to␈α⊂the␈α⊂system.␈α⊂ While␈α⊂it␈α∂is
␈↓ ↓H␈↓␈↓ αHpossible␈αto␈αrearrange␈αthe␈αoperating␈αsystem␈αso␈αthat␈αit␈αcan␈αhave␈αenough␈αspace␈αto␈αdescribe
␈↓ ↓H␈↓␈↓ αHall␈αthe␈αmemory␈αand␈αall␈αthe␈αdisks,␈αto␈αuse␈αmemory␈αin␈αthis␈αway␈αwill␈αcut␈αus␈αo≥␈αfrom␈αfuture
␈↓ ↓H␈↓␈↓ αHsoftware developments that require this space.

␈↓ ↓H␈↓␈↓ αH4.␈αThe␈αupgrade␈αto␈αthe␈α2060␈αseems␈αnecessary␈αin␈αthe␈αlong␈αrun.␈α If␈αit␈αis␈αto␈αbe␈αdone␈αat␈αall,␈α
it
␈↓ ↓H␈↓␈↓ αHshould␈α∞be␈α
concurrent␈α∞with␈α
the␈α∞memory␈α∞upgrade␈α
to␈α∞be␈α
certain␈α∞that␈α
the␈α∞new␈α∞memory␈α
is
␈↓ ↓H␈↓␈↓ αHcompatible with the 2060's tighter timing tolerances.

␈↓ ↓H␈↓The␈α
price␈α
for␈α
the␈α
upgrade␈α
to␈α
the␈α
2060␈α
is␈α
$50,000,␈α
including␈α
the␈α
new␈α
processor␈α
and␈α
a␈α
version␈α
of␈α
the
␈↓ ↓H␈↓operating␈α∂system␈α⊂that␈α∂supports␈α∂the␈α⊂new␈α∂processor␈α∂features␈α⊂and␈α∂additional␈α∂memory.␈α⊂ [All␈α∂prices
␈↓ ↓H␈↓mentioned herein are exclusive of sales tax, insurance, shipping, and installation.]

␈↓ ↓H␈↓THE MAJOR DIFFERENCE:  More memory

␈↓ ↓H␈↓Analysis␈α∂of␈α∞operating␈α∂data␈α∞shows␈α∂that␈α∞in␈α∂the␈α∞worst␈α∂cases␈α∞observed,␈α∂only␈α∞30-40%␈α∂of␈α∞the␈α∂CPU␈α∞is
␈↓ ↓H␈↓delivered␈α∂to␈α∂users.␈α∂ Another␈α⊂20%␈α∂goes␈α∂to␈α∂scheduling␈α⊂users,␈α∂and␈α∂the␈α∂remainder,␈α⊂sometimes␈α∂more
␈↓ ↓H␈↓than␈α∞40%␈α∞is␈α∞wasted.␈α∞ The␈α∞problem␈α∞of␈α∞vast␈α∞amounts␈α∞of␈α∞wasted␈α∞time␈α∞(technically␈α∞known␈α∞as␈α
"swap
␈↓ ↓H␈↓wait")␈αcan␈αbe␈αdealt␈αwith␈αmost␈αe≥ectively␈αby␈αadding␈αmore␈αmain␈αmemory.␈α We␈αexpect␈αthat␈αwith␈α512
␈↓ ↓H␈↓Kw␈α
additional␈α
memory␈α
the␈α
swap␈α∞wait␈α
time␈α
could␈α
be␈α
dropped␈α
to␈α∞5-10%␈α
in␈α
the␈α
worst␈α
case,␈α∞with␈α
a
␈↓ ↓H␈↓corresponding increase in useful work to the 65-75% range.

␈↓ ↓H␈↓We␈α∂have␈α⊂observed␈α∂similar␈α∂systems␈α⊂that␈α∂have␈α∂more␈α⊂main␈α∂memory.␈α∂ They␈α⊂perform␈α∂substantially
␈↓ ↓H␈↓better␈α∞under␈α∞heavy␈α∞load␈α∞than␈α∞LOTS␈α∞does.␈α
 Among␈α∞the␈α∞systems␈α∞we've␈α∞seen␈α∞are␈α∞the␈α
DECsystem-
␈↓ ↓H␈↓1090T␈αat␈αSRI␈αand␈αthe␈αKL10-based␈αsystem␈αat␈αArti≡cial␈αIntelligence␈αLaboratory.␈α SRI␈αhas␈α1024␈α
Kw
␈↓ ↓H␈↓of␈α
main␈α
memory,␈α
and␈α
runs␈α
essentially␈α
the␈αsame␈α
hardware␈α
and␈α
software␈α
as␈α
LOTS.␈α
 Their␈αsystem
␈↓ ↓H␈↓runs␈α∂100␈α∂interactive␈α∂users␈α∂and␈α∂performs␈α∂much␈α∂better␈α∂than␈α∂ours␈α∂does␈α∂when␈α∂we␈α∂are␈α∂running␈α∞60
␈↓ ↓H␈↓users.␈α The␈αprincipal␈αdi≥erence␈α
is␈αthat␈αSRI␈αhas␈α
twice␈αas␈αmuch␈αmemory␈α
as␈αwe␈αhave.␈α The␈α
Arti≡cial
␈↓ ↓H␈↓Intelligence␈αLaboratory␈αhas␈αrecently␈αadded␈α512␈αKw␈αto␈αtheir␈α384␈αKw␈αsystem.␈α Performance␈αis␈αnow
␈↓ ↓H␈↓substantially better than it was.

␈↓ ↓H␈↓There␈αare␈αtwo␈αvendors␈αof␈αadd-on␈αmemory.␈α
 DEC␈αhas␈αrecently␈αannounced␈αa␈αsolid-state␈αmemory␈α
for
␈↓ ↓H␈↓the␈α⊗DECsystem-20.␈α↔ To␈α⊗double␈α⊗the␈α↔existing␈α⊗memory,␈α⊗using␈α↔DEC's␈α⊗new␈α⊗memory␈α↔will␈α⊗cost
␈↓ ↓H␈↓approximately $196,000. This cost breaks down as follows:
␈↓ ↓H␈↓␈↓ αH756 Kw of MOS memory␈↓ λx$250,000
␈↓ ↓H␈↓␈↓ αH256 Kw of core, trade in␈↓ λx( 54,000)
␈↓ ↓H␈↓␈↓ αHTotal:␈↓ λx$196,000

␈↓ ↓H␈↓The␈α⊂present␈α⊂memory␈α⊂is␈α⊂512␈α⊂Kw␈α∂of␈α⊂core,␈α⊂in␈α⊂two␈α⊂256␈α⊂Kw␈α∂boxes.␈α⊂ To␈α⊂make␈α⊂room␈α⊂for␈α⊂the␈α∂new



␈↓ ↓H␈↓αProfessor G. Lieberman␈↓ ¬VFebruary 16, 1979␈↓ 
nPage 7␈↓ 

␈↓ ↓H␈↓memory,␈αone␈αbox␈αhas␈αto␈αbe␈αremoved.␈α The␈αtrade-in␈αallowance␈αon␈αthe␈αold␈αmemory␈αis␈α$54,000.␈α
 The
␈↓ ↓H␈↓new␈α∩memory␈α∪will␈α∩be␈α∩768␈α∪Kw:␈α∩256␈α∪Kw␈α∩replacing␈α∩the␈α∪core␈α∩that␈α∩was␈α∪removed,␈α∩plus␈α∪512␈α∩Kw
␈↓ ↓H␈↓additional␈α∞memory.␈α
 MOS␈α∞memory␈α
from␈α∞DEC␈α
cannot␈α∞be␈α
delivered␈α∞before␈α
January␈α∞1979.␈α
 MOS
␈↓ ↓H␈↓memory absolutely requires the upgrade to the 2060 processor.

␈↓ ↓H␈↓A␈αsecond␈α
plan␈αalso␈α
involves␈αDEC␈α
as␈αthe␈α
vendor,␈αbut␈α
is␈αunorthodox.␈α
 This␈αplan␈α
calls␈αfor␈α
us␈αto␈α
buy
␈↓ ↓H␈↓512␈α∂Kw␈α∂additional␈α∂core␈α∂memory␈α∂and␈α∂connect␈α∂it␈α∂to␈α∂the␈α∂existing␈α∂system.␈α∂ Engineers␈α∂at␈α∂DEC␈α∞are
␈↓ ↓H␈↓uncertain␈α∃about␈α∀whether␈α∃such␈α∀a␈α∃connection␈α∀would␈α∃work,␈α∀particularly␈α∃on␈α∀a␈α∃2060␈α∀processor.
␈↓ ↓H␈↓However,␈α∞DEC␈α∞is␈α∂willing␈α∞to␈α∞let␈α∞us␈α∂experiment␈α∞and␈α∞if␈α∞the␈α∂experiment␈α∞reveals␈α∞problems␈α∂we␈α∞may
␈↓ ↓H␈↓return␈αthe␈αmemory.␈α Also,␈αDEC␈αhas␈αagreed␈αto␈αlet␈αus␈αlease␈αthe␈αmemory␈αon␈αa␈αtrial␈αbasis.␈α The␈αexact
␈↓ ↓H␈↓details are somewhat complicated but the proposal is roughly:

␈↓ ↓H␈↓␈↓ αH512␈α⊗Kw␈α⊗memory␈α⊗and␈α⊗cabinet,␈α↔$140,000.␈α⊗ Lease,␈α⊗$7,000/month,␈α⊗of␈α⊗which␈α↔100%␈α⊗is
␈↓ ↓H␈↓␈↓ αHapplicable to purchase.  Maintenance while on lease is an additional charge.

␈↓ ↓H␈↓It␈αis␈αnot␈αnecessary␈αfor␈αus␈αto␈αupgrade␈αto␈αa␈α2060␈αprocessor␈αbefore␈αinstalling␈αthis␈αmemory.␈α However,
␈↓ ↓H␈↓there␈α⊃is␈α⊃a␈α⊃chance␈α⊃that␈α⊃the␈α⊃core␈α⊃memory␈α⊃will␈α⊂not␈α⊃work,␈α⊃or␈α⊃will␈α⊃work␈α⊃less␈α⊃well␈α⊃with␈α⊃the␈α⊂2060
␈↓ ↓H␈↓processor␈α
than␈α∞with␈α
the␈α∞2050␈α
we␈α
already␈α∞have.␈α
 It␈α∞would␈α
be␈α
prudent␈α∞for␈α
us␈α∞to␈α
try␈α∞the␈α
extended
␈↓ ↓H␈↓memory con≡guration with the 2060 before buying the memory.

␈↓ ↓H␈↓The␈α⊃third␈α⊂plan␈α⊃involves␈α⊂another␈α⊃vendor,␈α⊂Ampex.␈α⊃ They␈α⊂propose␈α⊃to␈α⊂add␈α⊃core␈α⊂memory␈α⊃to␈α⊂the
␈↓ ↓H␈↓existing␈α⊂system.␈α⊂ The␈α⊂price␈α⊂from␈α⊂Ampex␈α⊂for␈α∂512␈α⊂Kw␈α⊂of␈α⊂their␈α⊂core␈α⊂memory␈α⊂is␈α⊂$115,000.␈α∂ The
␈↓ ↓H␈↓Ampex␈α∩memory␈α∩can␈α∩also␈α∩be␈α∩had␈α∩on␈α⊃a␈α∩short-term␈α∩lease,␈α∩for␈α∩$5,000␈α∩monthly,␈α∩with␈α⊃$4,480/mo
␈↓ ↓H␈↓applicable␈αto␈αthe␈αpurchase␈αof␈αthe␈αmemory.␈α The␈αdisadvantages␈αof␈αgoing␈αwith␈αthe␈αAmpex␈αmemory
␈↓ ↓H␈↓are:

␈↓ ↓H␈↓␈↓ αH1.␈α⊃Potential␈α⊃hardware␈α⊃problems␈α⊃preventing␈α⊃the␈α⊃memory␈α⊃from␈α⊃working␈α⊃reliably.␈α⊃ (It's
␈↓ ↓H␈↓␈↓ αHimportant that we evaluate the performance of the memory on a 2060 system.)

␈↓ ↓H␈↓␈↓ αH2. Di≠culties in dealing with suppliers and maintenance in a multi-vendor environment.

␈↓ ↓H␈↓O≥setting␈α∃these␈α∀negative␈α∃considerations␈α∀is␈α∃the␈α∀price.␈α∃ Moreover,␈α∀engineers␈α∃at␈α∃Ampex␈α∀have
␈↓ ↓H␈↓demonstrated␈αthat␈α
their␈αmemory␈α
will␈αwork,␈α
while␈αthere␈α
is␈αsome␈α
doubt␈αthat␈α
the␈αDEC␈α
core␈αwill␈α
work
␈↓ ↓H␈↓in our con≡guration.  It currently looks as though going with Ampex is the best thing to do.

␈↓ ↓H␈↓ALSO NEEDED:  A second tape drive.

␈↓ ↓H␈↓The␈αpresent␈αtape␈αdrive␈αis␈αadequate␈αfor␈αmost␈αpurposes,␈αbut␈αwhen␈αLOTS␈αis␈αvery␈αbusy␈αit␈αis␈αdi≠cult
␈↓ ↓H␈↓for␈α
a␈α∞student␈α
to␈α
use␈α∞the␈α
drive␈α
because␈α∞it␈α
is␈α
tied␈α∞up,␈α
for␈α
many␈α∞hours␈α
at␈α
a␈α∞time,␈α
writing␈α∞our␈α
disk
␈↓ ↓H␈↓backup␈α∂tapes.␈α∂ Because␈α∂of␈α∂the␈α∂general␈α⊂slowness␈α∂of␈α∂the␈α∂system,␈α∂we␈α∂cannot␈α∂e≥ectively␈α⊂backup␈α∂all
␈↓ ↓H␈↓three␈αdisk␈αpacks;␈αthis␈αprevents␈αus␈αfrom␈αtaking␈α
full␈αadvantage␈αof␈αthe␈αthird␈αdisk.␈α The␈αtape␈αdrive␈α
is
␈↓ ↓H␈↓also␈α∪our␈α∩major␈α∪communication␈α∩path␈α∪to␈α∩DEC,␈α∪SRI␈α∩(with␈α∪whom␈α∩we␈α∪share␈α∩software)␈α∪and␈α∩the
␈↓ ↓H␈↓Graduate␈α
School␈α
of␈α
Business.␈α
 For␈α
our␈α
users,␈α
the␈α
tape␈α
drive␈α
is␈α
the␈α
main␈α
path␈α
to␈α
SCIP.␈α
 We␈α
are
␈↓ ↓H␈↓anxious␈α
that␈α
people␈α∞be␈α
able␈α
to␈α
take␈α∞advantage␈α
of␈α
both␈α
LOTS␈α∞and␈α
SCIP,␈α
in␈α
whatever␈α∞mix␈α
best
␈↓ ↓H␈↓suits␈αtheir␈αneeds.␈α Lack␈αof␈αan␈αavailable␈αtape␈αdrive␈αinhibits␈αpeople␈αfrom␈αutilizing␈αthe␈αbest␈αfeatures
␈↓ ↓H␈↓of each system.



␈↓ ↓H␈↓αProfessor G. Lieberman␈↓ ¬VFebruary 16, 1979␈↓ 
nPage 8␈↓ 


␈↓ ↓H␈↓To␈α∞some␈α∞extent,␈α
having␈α∞more␈α∞main␈α
memory␈α∞is␈α∞expected␈α
to␈α∞relieve␈α∞some␈α
contention␈α∞for␈α∞the␈α
tape
␈↓ ↓H␈↓drive␈α
by␈α
making␈α
the␈α
disk␈α
backups␈α
go␈α∞faster.␈α
 However,␈α
if␈α
we␈α
also␈α
expand␈α
to␈α
three␈α∞public␈α
disks,
␈↓ ↓H␈↓we'll be back in the same place as before, with respect to accessing the tape drive.

␈↓ ↓H␈↓The additional tape drive costs $15,000.

␈↓ ↓H␈↓IN SUMMARY:

␈↓ ↓H␈↓The␈α2060␈α
upgrade␈αis␈αnecessary␈α
in␈αthe␈α
long␈αrun␈αto␈α
be␈αcompatible␈α
with␈αDEC␈αsoftware␈α
developments
␈↓ ↓H␈↓and␈αto␈α
take␈αadvantage␈αof␈α
new␈αfeatures␈αas␈α
they␈αbecome␈α
available.␈α However,␈αif␈α
we␈αbuy␈αmemory,␈α
we
␈↓ ↓H␈↓need the processor upgrade concurrently, to be sure the memory will work.

␈↓ ↓H␈↓2060 Upgrade:␈↓ 
_$ 50,000

␈↓ ↓H␈↓Apart␈αfrom␈αan␈αadditional␈αentire␈αcomputer␈αsystem,␈αnothing␈αwill␈αbetter␈αhelp␈αLOTS␈αmeet␈αthe␈αneeds
␈↓ ↓H␈↓of␈αthe␈αuser␈αcommunity␈αthan␈αmore␈αmemory.␈α We␈αneed␈αto␈αdouble␈αthe␈αmemory.␈α We␈αhave␈αtwo␈αo≥ers
␈↓ ↓H␈↓meet the need by Fall quarter 1978.

␈↓ ↓H␈↓512 Kw additional memory from DEC, purchase:␈↓ 
_$140,000

␈↓ ↓H␈↓␈↓ αHshort␈α∞term␈α∞lease:␈α∂$7,000/month␈α∞plus␈α∞maintenance␈α∞100%␈α∂of␈α∞the␈α∞lease␈α∞price␈α∂applicable␈α∞to
␈↓ ↓H␈↓␈↓ αHpurchase after 3 months.

␈↓ ↓H␈↓512 Kw additional memory from Ampex, purchase:␈↓ 
_$115,000

␈↓ ↓H␈↓␈↓ αHshort␈α∞term␈α∞lease:␈α∞$5,000/month␈α
including␈α∞maintenance.␈α∞ $4,480/month␈α∞applicable␈α∞to␈α
the
␈↓ ↓H␈↓␈↓ αHpurchase␈α∩for␈α∩months␈α∩1␈α∪through␈α∩3.␈α∩ For␈α∩months␈α∩4␈α∪through␈α∩6,␈α∩60%␈α∩of␈α∪$4,480/mo␈α∩is
␈↓ ↓H␈↓␈↓ αHapplicable to the purchase.

␈↓ ↓H␈↓Purchase of an additional tape drive:␈↓ 
_$ 15,000

␈↓ ↓H␈↓Our recommendation:

␈↓ ↓H␈↓␈↓ αH1.  Purchase of the upgrade to the 2060 processor␈↓ 
_$ 50,000

␈↓ ↓H␈↓␈↓ αH2.  Purchase of an additional tape drive␈↓ 
_$ 15,000

␈↓ ↓H␈↓␈↓ αH3.␈α Lease␈αof␈αan␈αadditional␈α512␈αKw␈αmemory␈αfor␈α3␈αmonths,␈αincluding␈αmaintenance,␈αnot␈αto
␈↓ ↓H␈↓␈↓ αHexceed␈↓ 
_$ 22,500

␈↓ ↓H␈↓␈↓ αH4.␈α
 If␈α
the␈α
memory␈α
and␈α
system␈α
behavior␈α
is␈α
satisfactory,␈α
purchase␈α
of␈α
the␈αmemory,␈α
including
␈↓ ↓H␈↓␈↓ αHallowance for lease credits, not to exceed␈↓ 
_$120,000

␈↓ ↓H␈↓␈↓ αHTotal, exclusive of tax, shipping, insurance␈↓ 
_$207,500

␈↓ ↓H␈↓␈↓ αHEstimated maintenance cost of the above,␈↓ 
_$ 10,000/year