LT=JUL@75 17234841=PDT,33777000000000000

Date? 17 JUL 1975 1734=PDT

From RINDFLEISCH

Subject; COMMENTS ON WIPKE PROPOSAL

Tot LEDERBERG, LEVINTHAL, FEIGENBAUMs BUCHANAN, SMITH

The following are some comments on the Wipke proposal from the view
point of system loading, I have enly curserily jooked aver his
communications estimates and have not had time to verify his numbers
with the telephone company heres

There are two major areas af commenty $) the realiam of his estimates
and 2) the relation between his proposed project and cuprent system
 

9 <LEVINTHALDWIPKE, #1 MON 24*eJUL*75 9:40AM

use. His estimates of system loading includet

CPU time 22 = 65 hours/month
Connect time 430 hours/month
File space 8000 pages

These are to be compared to the following data for currently eetive
Projects (averages over April, May, and June)! .

CPU Hes, Conn Hes, File Pgs
DIALOG (4,63 164 5100
Rutgers 12,15 488 6210
DENDRAL - §1,68 (3375 19020 |
MYCIN 20,16 ; 4aq 4500

Protein Model 5,56 . £67 1220

There {8 a great deal of Uncertainty {n his estimates of CPU time
“Cfactor of 3), It seema clear that the high end cannot be correct,
for example as compared to the DENDRAL usage, This {8 borne out tn
some conversations with Peter Friedland about the relative demands
of Wipke's programs and say CONGEN, Wipke's code js all im FORTRAN
_@8, opposed to LISP and should place considerably Jess demand, 1
think there are a lot of uncertainties is Wipke's estimates for
legitimate reasons (he {a coming from a smal! KAe{0 [64K] under an
ancient DEC system to a large care KI“TENEX), FPriediand felt that much
time was spent in disk aecess for overlays. I am trying to Bin some
of these numbers down better, I. am trving to Peach Wipke to propose
that comparative runs be made with the "Model Builder” on his machine
and with the version running here © this may provide a seale for the
translation, It is not elear this can be done before Wipke leaves
Prineeton, .

If we take the lower_estimate of CPU usages his project would be
about of the seale of MYCIN, His statement that "mot al! of this time
Would be needed during prime shift” and that he “would be wilting

to assist the community {nm the solution to loading problems” suggest
that his work eould be accommodated at this fawer level of usage.

The terminal access time does not seem grossly out of reason, As I
_taid above, I can't confirm his estimates of I{ne charges,

‘His estimates of file requirements are highs at feast as I read the
rationale for his numbers. The 8000 page estimate seems to be based
On multiple copies of SECS (about 5 or 6 copies at 1200 pages each).
It seems like poor software contro! to have so many copies of sources,
rel files and binaries lying around) Data files I could understand
but these afe not mentioned. Thus I think his requirements for disk
Space may reflect the fact that he was On a ome man machine with lots
of file space and he expanded to fitl dt. It is hard to estimate the
Peal need for programs and data = perhaps 2 versions plus data at
around 3000°4000 pages would be enough, If his file meeds are {ndeed
as large as he says (40%. of a disk drive), then we will need to ask
his grant proposal te include funds to augment our file system
appropriately,
$ “LEVINTHALAWIPKE, 31 MON 2t@JUL#75 93:40AM

I will pass on other data as it becomes available,

Tom R,