The first semester of the WIYN queue observing experiment has just been completed and it's safe to say it was a learning experience for all involved. We are in the process of compiling observing statistics for the fall 1995 semester and will report them on our WWW page (http://www.noao.edu/wiyn/queue) by 1 March 1996. A few general remarks about the lessons we have learned during fall 1995, and how we plan to continue the experiment in the spring 1996 semester follow.
The technical triumph of WIYN has naturally led to an enthusiastic outpouring of observing proposals. Unfortunately, this has produced a level of oversubscription that is frustrating to all of us. For the spring 1996 observing semester, KPNO received 58 WIYN observing proposals, compared with 83 for the Mayall 4-m. However, KPNO is allocated only 40% of the observing time at WIYN by the WIYN Consortium. This translated into a roughly 2.5:1 WIYN requested/available nights ratio, but the equivalent dark-time oversubscription (roughly 4:1) is even worse.
The main goal of the WIYN queue observing experiment is to test empirically the hypothesis that in the face of a high oversubscription rate, the science throughput of WIYN can be maximized by executing the most highly ranked science programs first, completing datasets in a timely manner, allowing a larger range of program lengths, and matching the observing program to the observing conditions on an observation-by-observation basis. In fall 1995, we further held to the idea that once programs were initiated, high weight should be given to completing them. Moreover, because fall 1995 proposals were submitted before we could give users a good idea of observing overheads and instrument efficiencies, our goal was to give projects the actual time required to complete the TAC reviewed science program, rather than the TAC recommended time allocations.
These latter constraints naturally led to two outcomes that may or may not have been desirable. First, because good observing overhead estimates were not initially available, many programs took much longer (by factors of 2-4) to execute than estimated by the program PI. Since the highest TAC ranked programs were also fairly large projects, this effectively minimized the number of programs initiated at WIYN during fall 1995. Second, by insisting that the highest ranked program be completed first whenever possible, programs were often executed under better observing conditions than they actually required. This led, for example, to spectroscopic programs being executed on nights of very good seeing because they were more highly ranked by the TAC than more lower ranked imaging projects that required the good seeing to be completed successfully. At some level this is undesirable since it mitigates what should be one of the strengths of queue observing, namely the ability to take advantage of rarer observing conditions.
For spring 1996, we have fine-tuned the experiment in several ways to increase the number of programs serviced and to better match observing programs to observing conditions. First, larger survey projects that appeared to require more immediate PI interaction were "classically" scheduled; i.e., they were given specific nights. Often such projects do not require 100% completion to answer their science question. By scheduling them classically, the burden of deciding which fields to observe in what order is shifted to the PIs, while allowing the TAC to re-evaluate the program on a semester-by-semeter basis if the program is not completed. Roughly half of dark time and a third of bright time were scheduled classically in spring 1996. The amount of fall 1996 time scheduled classically will be determined after the TAC reviews fall 1996 proposals. This change from 100% queue scheduling to a mix of "classical" and queue is result of our continuing efforts to maximize the quality and quantity of scientific results from this telescope. This change also reflects many of the views expressed about the implementation of queue scheduling that were written to us in the NOAO electronic forum.
Second, programs scheduled in the queue will be held to their TAC recommended time allocation. Realistic observing overheads and instrument throughput have now been widely enough disseminated that proposers appear to be making good estimates of total program execution time. This constraint should increase the number of programs initiated in spring 1996. Prospective WIYN proposers should see the WIYN queue program Web page (http://www.noao.edu/wiyn/queue) for further information about estimating program execution times.
Third, we have reduced the initial queue pool. At the start of fall 1995, we entered roughly two-thirds of the TAC reviewed programs into the queue pool because we wanted to be assured of having something to observe under all reasonable observing conditions at all available celestial coordinates. This meant, however, that the ratio of "approved" queue time to available queue time was roughly 4:1 before weather, better observing overhead estimates, and facility downtime is considered. With those factors accounted for, the effective ratio by the end of fall 1995 was closer to 6:1. As a consequence, many users who were told they were in the queue never received any WIYN data and were naturally very frustrated. Rather than disappoint people at the end of the semester, it was decided to raise the initial TAC grade threshold for getting into the queue, give more users a definitive "no" answer up-front, and aim for a approved/available ratio of roughly 2:1. Note that we still have to over-schedule the queue to assure that we can span sufficient observing condition parameter space during the entire semester. Also, the pool must be large enough to cover the situation where all nights available to WIYN have favorable observing conditions and no equipment failures.
Lastly, we will relax the "100% completion" goal in favor of more optimal matching of observing conditions to observing program. Effectively, this means that on nights with good seeing we will execute imaging programs that require good seeing, and on photometric nights we will execute programs that require photometric conditions, rather than just continuing to execute a higher ranked spectroscopic program. There is at present no set rule as to how far "down" in the TAC gradelist we will go to acquire a program to fit the current conditions; this is clearly a function of the particular program mix. However, we do not anticipate deviating from the TAC ordered list by more than a few tenths of a grade point. This change will increase the number of programs that get some data, while lowering the number of programs that are essentially completed.
During spring 1996, we will attempt to issue more timely queue progress reports via the WIYN queue program Web page (see the URL above). Furthermore, we will continue to answer any questions sent to wiynq@noao.edu in as timely a manner as possible, especially questions related to programs being executed at the time we receive the question. Of course, we remain committed to delivering the highest possible quality data.
We appreciate your continuing patience as this experiment evolves. Please continue to send us your feedback; we are listening and implementing changes to make this a more successful and satisfying program.
Dave Silva (dsilva@noao.edu), Dave De Young
Back to the WIYN Observatory Home Page
Back to the WIYN Observing Program Home Page
Last Updated: 3/4/96, dsilva@noao.edu