Duty-Cycle Metrics
Introduction
This is a very brief introduction to the Duty-Cycle Metrics (hereafter "Metrics")
program. The program aims to quantify how efficiently we provide observing time
to the observers by measuring what fraction of a night is spent with the science
detector recording photons from the sky. More details can be found on the Metrics
home page.
The program extracts or calculates various relevant "events" relating
to observing. These include start and stop times for images and spectra, guiding
events, slewing events, etc. The sequence of these events then allows us to
classify various time periods during the night as one of several observing "activities":
slewing, science, calibration, focusing, etc.
The nightly
timelineWeb page shows the events and activities for each night on each
telescope. Menus and links at the top of each page allow you to navigate to
other nights, or even to the nightlog summary for that night. You can select
different start and stop times for the graphics to zoom in on a particularly
interesting time period.
The top graphic shows the event timeline. Vertical bars show 12° twilights.
The time between 12° twilights is the formal definition of the "night."
No statistics are performed outside of this time frame.
The second graphic shows the translation of the event sequences into activities.
There are some interesting effects you can see here, for example the reclassification
of some images into "fine acquisition," particularly during LRIS slitmask
observing. (See Keck I for Jan.
4, 2003 for an example.) Also, if no science data is taken on a target that
shows a focus activity, the entire time during which the telescope was pointed
at that star is classified as focus, not just the time during which the focus
script was running. (The logic here being that if the OA did not have to focus,
they would not have pointed at that star.) It is from these activities that
the statistics are calculated.
Fig. 1. Events on K1 from UT 9:00 to 11:00 on 2003 Jan. 4. Note the narrow
LRIS-B exposures in the top line just after 10:00. These are images taken
in order to align a slitmask. At around 10:45 longer exposures begin. These
are the slitmask spectra.
Fig. 2. Activities from the same time period as Fig. 1. Note that the slitmask
alignment images do not show up as science. They have been reclassified
as fine acquisition.
The third pie chart graphic shows the fraction of time spent in the various
observing activities. The green "slice" is the important one that
we want to maximize: the total time spent with the instrument's science shutter
open and collecting data.
Caveats
There are some caveats to the system that you should keep in mind:
- Some events mask others. For example, weather masks everything. Engineering
masks everything except weather. (We are trying to measure normal observing,
not engineering.) Faults mask everything except weather and engineering. And
so on...
- Not all events are available. For example, early DEIMOS data were
removed from disk onto tape before the metrics software was prepared to import
their FITS header information. Sometimes various pieces of software are not
working, and the events contained in their log files are hence not available.
We are striving to get at least 2002 and later accurate, and have DCS log
files available for 2001 as well.
- Human input: Some of the event information, notably the times of
faults, is derived from entries in the nightlog made by the OAs. Engineering
and weather events are also currently taken from the nightlog, although this
is a recent (early 2003) invention. Other engineering and weather time loss
for 2002 and 2003 was estimated and entered manually, often with help from
the OAs' nightlog summaries, sometimes with help from the event plots. Some
nightlog summaries report time lost to weather that is a rough estimate of
the throughput loss over the entire night because of semi-opaque clouds.
These losses are not included in the metrics events. (The logic here is that
while they affect the rate at which photons are collected, they do not substantially
affect the average duty-cycle of observing.)
- NIRSPEC and NIRC-2: We are still working on the algorithm for the
science exposures, so they are not 100% accurate.
- AO: Guiding information is available only in the AO log files, and
metrics information is not available from these yet.
- Idle time: It would be of interest to know when nothing at all is
happening, when it should be. However, it is very difficult to know this,
using only log files.
- Fine and coarse acquisition: Because of the caveat about "idle
time" above, the activity definitions for acquisition absorb idle
time. Fine acquisition is functionally defined by guiding being on, while
coarse acquisition is defined as guiding being off. If you see a large amount
of acquisition time, it could be because there are missing science events,
or some other problem with the event information, or a great deal of "idle
time."
- Instrument overheads: There are two types of overheads we would like
to track: detector readout overheads and instrument reconfiguration overheads.
The former are, in some cases, not directly logged and rather tricky to calculate.
They have not yet been included for all instruments (LWS is an exception).
Instrument reconfiguration times are in many cases not recorded in instrument
log files. To address this issue will require making changes to the C code
that runs the instruments.