|
|||
| Joint Astronomy Centre |
|
|
|
| Show document only |
A Note on Observing OverheadsIn the widely-circulated report of the most recent meeting of the JCMT Advisory Panel (on June 12) it is stated that"...overheads on typical heterodyne observations have decreased from 40% to 20%". These numbers were provided by a JCMT Staff Scientist very soon after the new control computer was installed; in fact they refer to 'wasted' time as defined here. Since prospective observers may be tempted to base their time requirements on this statement some clarification is in order. In the general sense implied, for most line observations, such a major decrease is NOT the case, although there has been some improvement over the past year due to upgrades in hardware and a considerable software effort. The principal improvement has in fact come from the introduction of raster ('on-the-fly') spectral line mapping, and users should seriously consider using this mode where appropriate. Below I consider some specific common examples of observing overheads, and comment on the overall efficiency of observing.What I have done is look through some recent observing sessions, mostly those with which I have been closely involved. Generally, I have used sequences of scans which have been more or less continuous, that is, there has been essentially no dead time due to uncertainty on the part of the observer or some other delay. Usually in such cases, an ICL procedure has been active, or the Telescope Operator has typed ahead to allow several observations to proceed without additional intervention. In such cases the dead time does not exceed a few seconds. I calculate the elapsed time (Te) for an average scan to complete in such a sequence, and any software/hardware overhead in starting a scan is naturally included. The total integration time (Ti) requested is known from the observing log (e.g. the JOURNAL program) and the on-source (signal phase) time (Ts) can be derived knowing the observing mode. Then I derive the overhead (Over), defined as (Te - Ti)/Ti, the on-source efficiency (OSE), given by Ts/Te, and the 'wasted' time, defined as (Te - Ti)/Te. 'Wasted' time is the time when the system is not recording data of any kind.
Case Rx Mode sec/cyc Te Ti Ts Over OSE wasted NotesMost of the entries in this table are self-explanatory, particularly for those with some experience of observing with the JCMT. PSSW, BMSW, FRSW are position-, beam- and frequency-switching respectively. 'Raster' refers to the 'on-the-fly' continuous scanning mode. Particulars relating to the various examples are given in the last column. 'With' and without ('w/o') cal(ibration) refers to whether an off- source calibration is done prior to the scan or not. Generally the latter is the norm, now that continuous calibration has been enabled with the DAS. In preparing proposals the prospective user will be most interested in the 'overhead' column. If you have calculated the total integration time (Ti; sig. + ref.) then a typical observation will take that much percentage longer to actually perform. The OSE column tells you how much of the elapsed time is actually spent on source. Most of the common modes split the integration time 50-50; the rastering and frequency-switching modes spend most, or all, of the time on-source, respectively. For UKT14 dual-beam raster maps, included for comparison, both beams are included in the map, and so almost no time is lost. The most inefficient modes are position-switching (the actual overhead depends somewhat on the distance to the reference), and C2 observations with intervening calibration (the calibration action is very slow in the hardware). The on-source efficiency is also low for these modes. The modes with the lowest overhead, and the highest on-source fractional times are frequency-switching and rastering. With the latter note that for a spectral line raster each row of n points, t secs per point, is accompanied by a reference spectrum of length tÖ(n) secs. Three instances of case 5 are given for different dates; similar obs. were done on the same source. Case 5(a) appears to have significantly higher overhead than either of the other two. Probably this is due to software development of the procedure in the interim. However, there is no difference between cases 5(b) and 5(c). This may be interesting, in that the later data were taken with a faster telescope control computer (VAX 4000/90 vs 4000/60). This table illustrates the point that the spectral line overhead is anywhere in the region between about 13% and almost 50%. 20% is not a representative number. My experiments, incomplete as they are, do however suggest that for these standard modes, one should be able to scale the results to arrive at sensible estimates of elapsed time. The position- and beam-switching and raster modes are independent of the receiver being used. They also say that there has been a significant improvement in overhead since the last version of the Guide for the Prospective User was made available (25th January). Finally, it is of interest not just to calculate the efficiency of individual scans, but also of a complete observing shift. I have modified the JOURNAL routine to do this (new version available soon). What it reports is that, for the few shifts I have examined in developing the program, the telescope is gathering photons for typically 55-65% of the time elapsed between the beginning of the first scan and the end of the last. This includes pointing and focus measurements. However, the program has no knowledge of what happened before the first scan started, although it does make a guess as to when the last scan ended. If one adds on the 15-30 minutes taken to set up the system, open doors and roof, tune receivers and so on, then one arrives at the estimate of the total time needed for a given program is approaching twice that of the total integration time needed. Experienced observers have felt this in their bones for a long time, but now I can prove it. Henry Matthews, JAC Information Coordinator
|