File name Alto_Time_Standard_Feb78.pdfInter- Office Memorandum
To PARC/SDD Date February 26, 1978
From Ed Taft Location Palo Alto
Subject Alto Time Standard Organization PARC/CSL
(Edition 3)
XEROX
Filed on: AltoTime.Ears, .Press, AltoTime.Bravo
I am resurrecting a proposal I made nearly two years ago to change the Alto's internal date
and time standard. The need for this change has become more pressing since that time, so I
would like to see it accomplished as quickly as possible.
A number of software systems will be affected by the change. This memo attempts to
enumerate them, but I have undoubtedly missed some (particularly ones in the Mesa and
Smalltalk worlds). Fortunately, the consequence of failing to convert immediately is not
catastrophic.
The last section of this memo sets forth a schedule for converting from the old standard to
the new. My intention is that the conversion be relatively smooth and that old programs
continue to work until such time as support for the old standard is revoked.
Change since Edition 1: The internal representation of a time zone is entirely different so
as to accomodate fractions of an hour and other oddities.
Change since Edition 2: Current status.
Present Standard
The current Alto time standard is a 32- bit integer denoting the number of seconds since
midnight, January 1, 1901, local time (with Daylight Savings Time already applied if it's that
time of the year). This standard has two important difficulties.
First, it is location- dependent: a given 32- bit integer represents a different absolute time
depending on the time zone within which it is interpreted. With the proliferation of Altos
across many time zones and the certainty that they will eventually be able to communicate
with each other, it becomes desirable that a time standard be location- independent. Given
this, a host in one time zone may obtain the current date and time (as a 32- bit number)
from a host in another time zone, without either host having to know the location of the
other. This capability is needed now because Webster has just been connected to our present Parc/SDD/XEOS
inter- network.
Second, the present time standard is not monotonic: it jumps forward an hour in April and
backward an hour in October so as to conform to daylight and standard time. Hence an
Alto's clock cannot be maintained correctly simply by counting milliseconds, and for this
reason it is not maintained correctly at present. Furthermore, there is a two- hour stretch of
absolute time every October within which each 32- bit num |