[daip] Re: Large experiments, VBGLU, etc.

[Jon Romney]

Hi Jim,

I fear this is an example of the communications gaps that you
mentioned in the recent all-hands meeting.

Looking ahead to the stage when we have added just one more
pair of Mark 5 record/playback units, and become able to stop
all recording on tape at NRAO instruments, I realized how big
a perturbation even occasional tape recordings were going to
be.  I discussed this with various EVN people, since I see the
global observations as the main problem, rather than the RDVs.
(As i recall, you warned the RDV organizers that we wouldn't
be able to correlate the large arrays they had been using, and
they said they understood and would reduce them.)  I hoped
this might also lead to a natural solution for the HSA obser-
vations, which I consider more important to us but which are a
less extreme case, and I believe this has happened.

I concluded that the multiple-station-subset correlation of
large observations would be the best solution if we could im-
plement it easily.  It would free us, and our users, from the
enormous distractions of the tape recording system.  It's im-
portant to do so as soon as possible for two reasons:

1)  To find out, for the first time, whether our current best
guess as to the Mark 5 media requirements is right, so we can
start including any necessary corrections in our planning.

2)  To begin to exploit the advantages of Mark-5-only opera-
tion.  This was one of the principal motivations for making
the change to Mark 5 in the first place.  These advantages
include:

     a)  Vastly superior data recovery from the recoding media,
     to the extent that manual flagging can become a thing of
     the past.

     b)  Freedom from tape-related concerns in scheduling and,
     especially, from gaps for tape changes at EVN stations.

     c)  Unattended correlation, resulting (eventually) in
     greater throughput for the VLBA.

Recording on tape periodically disrupts most of these goals,
and gives our users inferior results that they do not want.

I presented these thoughts in a long talk at the February
VLBA Test meeting.  Various concerns and additional points
were raised, but overall I felt everyone present agreed with
my suggestion.  In that sense, I would say "we" planned to
continue exploring that approach.  I realize you couldn't be
there yourself, but I did send you a copy of my presentation
to try to keep you informed.  I also gave a two-minute update
on further developments at the beginning of the March Test
meeting.

My goal in this, in everything I do, is to give our users the
the best VLBA instrument we can under the circumstances.
User-friendliness is important, but is only one aspect of
that.  I believe the advantages cited above (two of which
directly enhance user friendliness as well) far outweigh any
inconvenience of the multiple-station-subset correlations.

With all due respect, I have to say I feel the "misery" you
attributed to having to use VBMRG (not VBGLU) is a major
exaggeration.  The task has hardly any adverbs that have to
be set, and seems to run quickly.  Similarly, the preceding
and following sorts are easy to set up.  We don't seem to
consider the many other cases where sorts and special tasks
are required in other areas of AIPS processing to be user-
unfriendly, so why here?

Further, please consider the following mitigating factors:

1)  The whole process could be invisible to users who work
with pipeline results if, as I have suggested, we include
VBMRG in the pipeline where it is required.

2)  If the sorts are considered a problem, perhaps VBMRG
itself could run them.  I realize this may be un-AIPS-like.

3)  Whatever misery is caused would be self-limiting in the
global observations that should be the most common use of
this approach.  The policies for globals give proposers the
right to select which correlator will process their observa-
tion, as I have been informed.  Those for whom the multiple-
station-subset correlation will be too miserable can go else-
where, leaving us more time to concentrate on the VLBA-only,
NRAO-only, and HSA observations that best show what the VLBA
can do.

We have not started doing any multiple-station-subset correl-
ation yet; only tests have been carried out and/or planned.
I have invested a major effort in this exploration already,
but it's not too late to abandon it.  It is, however, almost
too late not to talk with the EVN about it more seriously,
since I planned to bring it up at the TOG meeting this week.

After considering your points, and my own above, I still be-
lieve this approach optimizes (1) progress toward our Mark 5
goals; (2) operational efficiency; and (3) user friendliness.
I reconfirm my recommendation that we complete the develop-
ment of it.

Jon


Jim Ulvestad wrote:
> Hi all,
> 
> On the VLBA, why are we correlating projects (if we are) with
> 13 disk stations, for our 11 disk playbacks.  This makes life
> miserable for the users.  Shouldn't we be recording only 11
> stations on disk in that case, as I thought we had planned,
> and a couple stations on tape?
> 
> Making a large number of users use VBGLU is not conducive
> to making the VLBA more user friendly.  For 20-station
> RDVs, the geodetic folks can buy us playbacks if they want
> to get all 20 stations.  If we can get to 14-station support for
> HSA, I'd rather spend another $100K keeping St. Croix from
> collapsing than spend it on playbacks, and I bet the RDV people
> sure would like to have SC as well.
> 
> I'm copying this to daip because of the work that has been done
> on VBGLU and VBMRG recently.
> 
> Jim
>