Документ взят из кэша поисковой машины. Адрес оригинального документа : http://www.stsci.edu/ops/stories/10.text Дата изменения: Tue Jul 13 19:30:41 1999 Дата индексирования: Sun Apr 10 14:27:30 2016 Кодировка: Поисковые слова: images

Deliverance from the Overlords of Calendar Inefficiency

During the first few years of HST Operations, the Calendar Builders were
enslaved by the Science Planning Branch (SPB) to feed the hungry spacecraft
with a timeline of observations every week. The Calendar Builders had to
follow laws spelled out in documents called orderforms that were written by
an SPB Overlord each week. Orderforms contained only a small subset of the
hundreds of observations, a.k.a. Scheduling Units (SUs), that could schedule on
a given calendar, and each SU had a priority rating between 1-100 based upon...
apparently (INT(RND(1) * 100) + 1. If an SU that appeared on an orderform was
not scheduled, a Builder would have to write a humble explanation explaining
why for each one. Many times the unschedulable observation was physically
impossible to schedule due to orbital constraints and target location; other
times an observation would fail to schedule because it would only schedule
where a "higher priority" observation was already scheduled.
Sometimes, SPB would send a new version of the orderform with replacement
SUs, but often snapshots would have to be used. The GIMME_SNAPS "tool", a
script created by one of the Calendar Builders, provided the only control a
Builder could have over what got scheduled during a week. Mostly because of
this highly iterative process, a Calendar Builder had to start building seven
weeks prior to uploading the schedule to the spacecraft. It took four weeks
to have a calendar ready to send to GSFC, of which the entire first week was
spent preparing SUs that appeared on the orderform.
In 1993, the Solar Array Drive Electronics (SADE) had difficulties. To
ensure the mechanism would last until Servicing Mission 1 at the end of the
year, the number of solar array movements per week had to be minimized.
For many months, the solar arrays could not be turned more than three times per
week. This meant that the spacecraft did not have the freedom to roll to
any observation in the sky not occulted by the earth, sun, or moon. The SPB
Overlords tried to present appropriate observations through their orderforms,
but many of the SUs could not meet the requirements. Those that met the
requirements were horribly inefficient. Finally, the Overlords gave up, and
decided to let the Calendar Builders find suitable SUs.
The Calendar Builders were given a line-of-sight tool from their software
developers, and they meticulously used it on every observation available
in the scheduling pool. The resourceful Builders quickly created a script
to sort the observations by the way they supported the legal SADE roll angles.
Despite the tremendous constraints, these calendars turned out to be over 40%
efficient, a marked improvement over the 28-33% efficiency of the past 2 years.
These improvements impressed those more powerful than the Overlords, and
the Calendar Builders were granted independence from their oppressors. The
Builders were now solely responsible for making sure that all observations were
scheduled before the end of an observing cycle in accordance with all the
special requirements imposed by the proposers. It seemed like a undaunting
task considering that there were thousands of SUs in the scheduling pool. At
first, it still took the Builders an entire week to create a list of SUs and
prepare them for use in the scheduling process, but they quickly grew tired of
this and made some serious improvements. The GIMME_SNAPS tool was beefed up to
extract any SU in the scheduling pool that had timing windows intersecting the
week. A tool called SUSORT was developed to quickly determine which SUs had
windows that did not go beyond the building week and assigning them the highest
priority. The software developers created a program called LREPORT to help the
Builders make sure that no members of a linked set of observations would become
"orphaned." These improvements alone dropped the calendar building cycle from
seven weeks to six weeks.
A couple years later, the Calendar Builders became part of PRESTO.
Further enhancements were made to their process. The software developers
created the CALOPT program, which was able to find every place an SU was
schedulable during a week. The Long Range Plan was formed to help identify
places where unschedulable zones could be hidden and where the Continuous
Viewing Zones (CVZ) could be optimally used. The calendar building cycle
shrank to four weeks, then down to three weeks where it stands today. Calendar
efficiency has always been on the rise and is now consistently over 50% each
week. Since the Builders became unshackled from orderforms, calendars have
become nearly twice as efficient and are being built in less than half the time.