 News |
FAQ
|
| Search
|
Home
|
| Getting Started | Documentation | Glish | Learn More | Programming | Contact Us |
|
| Version 1.9 Build 1367 |
|
T.J. Cornwell, NRAO
2000 May 1
Development now proceeds via a 6 month cycle, the details of which were given in the last report. A development plan for this cycle was drawn up and circulated (AIPS++ Note 226. Reporting is also based on the development cycle.
In this document, we report on the work accomplished in this development cycle. The major accomplishment this cycle was the issuing and support of the first public release (version 1.2), and the preparation of the second public release (version 1.3). In addition, we continue to support other activities such as the GBT commissioning, and the HIA ACSIS development.
Jim Braatz, Joe McMullin and Bob Garwood continue to focus on support of the GBT and development of the dish environment for general single dish analysis in aips++. There continue to be regular visits by Garwood and McMullin to Green Bank and Braatz to Charlottesville. Braatz has primary responsibility for GBT support while McMullin has primary responsibility for dish development. The group also supports the efforts by the Arecibo staff to reduce Arecibo data through dish/aips++.
Users in Green Bank have utilized AIPS++ for a number of projects including radiative transfer code, analyzing spectral processor data, displaying GBT encoder data, and developing GBT user interface tools. A number of utility functions were added to the GBT commissioning tool in response to user feedback. Braatz presented a paper on the aips++ commissioning tool at the January AAS meeting.
Garwood continued to spend a fair amount of effort on the GBT data filler. This takes the various GBT FITS files produced during an observation and fills them into an AIPS++ MeasurementSet. The helper classes written during this period should make maintenance much simpler and they should prove useful for other fillers.
The non-GBT focus this cycle was primarily on responding to reported defects and enhancement requests from users of dish. We gained a number of users when members of the Arecibo staff began actively using AIPS++ and dish. McMullin visited Arecibo in February 2000 to talk to them about using dish and to listen to them describe their needs. User feedback from Arecibo users and Green Bank users will guide our development plan for the next development cycle.
McMullin gave the second in a planned series of talks on AIPS++ to the NRAO-CV staff. This talk was also carried on the NRAO video conferencing system where it was seen by staff in Green Bank and Socorro. This talk covered the basics of using dish.
Priorities at present include: i) wider scientific use of the package; ii) broadening of the existing thin-path capabilities for connected-element arrays; iii) addition of thin-path processing for VLBI; iv) time-critical local priorities (e.g. TMS commissioning), and v) basic automated and real-time imaging. Actions this cycle in synthesis have been scheduled in keeping with these priorities.
Broadening of thin-path capabilities has occurred in several areas. These include: i) the completion of the infrastructure required to fully implement the MS v2.0 data format (M. Wieringa); ii) the addition of new or expanded features in the VLA (R. Marson) and ATCA fillers (M. Wieringa and F. Badia); iii) higher-level integration of calibration and imaging facilities (A. Kemball), and iv) upgrade of support for source component models, including projection (R. Marson) The MS v2.0 infrastructure has been checked-in to the system, to allow time to migrate data fillers and converters to the new standard, for a switch-over date in April 2000, as soon after issuing the second public release as practical. This format has resulted from a united effort throughout the consortium to provide a generic interferometric and single-dish data format.
As part of the overall thin-path strategy, advanced development has continued to be scheduled in targeted areas, primarily to prototype or complete new data reduction capabilities within the current synthesis framework. As part of this effort, work has continued in mosaicing development (M. Holdaway/T. Cornwell), 3-D/wide-field imaging (K. Golap/T. Cornwell) and single-dish/OTF gridding within the standard synthesis framework (T. Cornwell/B. Garwood). Planning has been completed for development of a standard holography reduction tool to extend existing tools for commissioning the GBT holography back-end (J. Braatz). A significant effort has also been made in extending the current simulation capabilities (M. Holdaway), both to allow testing of current synthesis capabilities, but also to meet simulation needs for ALMA and SKA. In mosaicing, work has continued in completing optimal mosaicing weighting, MEM and multi-resolution deconvolution and general testing. Further advanced development in synthesis is described in the parallelization section elsewhere in this report.
A new focus this cycle has been in the area of pipeline development, here defined to cover both automated off-line reduction as well as real-time imaging. Both these applications, in keeping with standard interactive reduction, use many common data reduction components from the synthesis infrastructure. The maturity of the synthesis library has allowed these efforts to start at several consortium telescopes, but there are significant gains to be made in sharing and coordinating common infrastructure in this area, as well as in expedited completion of the missing core synthesis capabilities. A meeting on pipeline development was held in January, 2000 to review scientific requirements in this area and coordinate common development. This was a successful meeting, and has led to the establishment of a coordination group and mechanism for those parts of development which can be shared. At present, common scientific requirements and infrastructure components are being distilled from discussions and presentations at the meeting.
Oleg Smirnov joined the AIPS++ group as a developer at NFRA during this development cycle. His first assignment has been the implementation of an AIPS++ interface to the PIM ionosphere model, and integration of an initial version in the existing calibration infrastructure. Signficant progress in this area was achieved during a successful visit Oleg made to the AOC early this year.
In the closing stages of this development cycle, a significant effort has been made in release testing and defect correction, in common with the rest of the project.
Future work in synthesis in 2000 will follow the overall priorities enumerated above. This will continue the expansion of the thin-path capabilities as described, in addition to the new efforts in VLBI and pipeline development. The area of VLBI has developed slower than desired, mostly due to personnel difficulties. However, these problems are now alleviated, and we expect significant progress in this area in the next development cycle.
The immediate objective remains the completion of several key data reduction projects to demonstrate the scientific usefulness of the parallelization capabilities. These projects selected for this effort represent mosaicing, 3-D/wide-field imaging, and large, multi-epoch VLBI observations. Parallelization development tasks this cycle have been chosen primarily to support this effort.
A document requesting computer policy accommodations, primarily in the area of persistent (non-batch) disk storage, was prepared by D. Roberts and submitted to the policy committee at NCSA. This was followed by a time allocation request to cover the key projects specified above. We have also agreed the operational procedures to be used in reducing the key projects as part of this effort.
W. Young completed a port of AIPS++ to full 64-bit operation under the SGI native compiler, to remove the existing 32-bit limit of 1 GB on memory allocation, which is anomalously low compared to the available memory on large high-performance systems, and has restricted key project processing in mosaicing in the past. A change proposal and AIPS++ document (Note 227) has been prepared to unify 64-bit conventions within AIPS++ (W. Young / A. Kemball). D. Roberts has implemented a script for the M33 mosaicing key project, to ensure that large imaging using this infrastructure is working correctly.
The parallelization infrastructure classes were updated to their next major revision to include greater support for multi-algorithm parallelization (A. Kemball / W. Young). This is required for 3-D/wide-field parallelization. K. Golap has continued work in this area, both verifying the correctness of the serial 3-D imaging with T. Cornwell, but also in implementing prototype parallelized code in this area. This has overlapped with continuing parallel I/O instrumentation work (D. Roberts) using the Pablo library.
Parallel I/O work were presented as a poster at ADASS'99 by D. Roberts. Trials of statement-level parallelization have also continued using OpenMP (D. Roberts) as part of the infrastructure development.
The NT port has continued in this cycle (P. Cortes), and is making good progress. A full installation on the Linux supercluster at UNM in Albuquerque was completed this cycle (W. Young), and has be used for cluster profiling of AIPS++ applications under Linux. This will be compared with NT cluster performance once the NT port is complete.
Much of the recent work on Glish involves fixing memory leaks which arise from the way reference counting works. With reference counting, each object contains a count of the number of places which refer to it. Each time another object established a reference to an object, the reference count of that object is incremented. Each time a reference goes away the referenced object's count is decremented. This works fine as long as there are no cycles in the reference graph, but as soon as a cycle arises all of the objects participating in the cycle are leaked.
The first attempted fix for this problem was to introduce garbage collection. The introduction of garbage collection in a limited manner is the way reference count cycles are typically addressed. In our case, we added a freely available garbage collector, and while this worked very well, it doubled execution time. This added overhead was unacceptable.
Since leaks in Glish are limited to function declarations within a function invocation, most of the Glish work during the past few month has been devoted to understanding how these cycles are created, and adding extra mechanics to analyze the results of function invocations and attach enough meta-information to the results to ensure that when the results of the function are no longer used all of the memory allocated as part of the invocation will be deleted.
This proved to be a fair amount of work because Glish is very flexible, and because the source of the problem was at the core of the Glish mechanics. This work is nearing completion and will be integrated and tested during the next development cycle.
Aside from this problem, most of the other defects were of a minor nature and were fixed in a day.
The only substantial change was to add handling of Comets and other solar system bodies to Directions.
Improvements were made to the user interface. The tool manager was simplified, and new entry widgets for a number of Measures were developed. The TableMeasures classes developed by Michael Haller have been improved and changed according to the MS2 needs. All classes have been copped and moved to the aips tree.
Wim Brouw developed a very general set of C++ classes and Glish tools for fitting. These will now form the basis of most fitting in AIPS++.
About 15 classes in the Lattices module have been reviewed by Peter Barnes and changed according to his comments. This first chunk of Lattices classes is about to be moved to the aips tree. For this purpose a few common classes have been moved from Lattices to Arrays, otherwise a mutual dependency would exist.
Several bugs have been fixed in the Arrays, Tables, and Lattices module. Table locking/synchronization has been improved. The precedence of the power operator is standardized in TaQL and LEL, so -22 = - 4 and not 4. The median function has been added to the Lattice Expression Language, so one can get the median of an image (section). Ms2uvfits now correctly handles multi-source measurementsets.
A class ScaledComplexData has been developed to be able to store Complex data points as 2 16-bit integers to reduce file size.
The cookbook was expanded considerably and renamed to ``Getting Results in AIPS++''. Other than that and the normal additions for new tools, only minor changes and revisions were made.
The major news is that Jodrell Bank Observatory/MERLIN rejoined the AIPS++ consortium. Barry Maguire was hired by JBO and at the time of writing is in Socorro for training.
A number of staff changes have occurred, with the consequent need for reallocation of tasks, and retraining. Staff difficulties have particularly affected the synthesis development, slowing down progress substantially.
We have started planning for an AIPS++ User Group to replace the long-defunct STAG. We expect to hold the first meeting in the late summer. The list of invitees is being constructed now. The emphasis will be on active users of AIPS++ as judged from defect reports, email logs, etc.
The AIPS++ user meetings at the AOC have continued with the goal of preserving and expanding the existing scientific users within NRAO. The AOC management expedited the formation of an active testers group at the AOC, led by Steve Myers. Outreach efforts have also taken place at most consortium sites to achieve the same purpose. A joint visit was undertaken in November 1999 by T. Cornwell and A. Kemball to several sites in Europe to publicize AIPS++ to a wider international community. Additional visits of this nature, both within the US and elsewhere, will be scheduled in 2000.
We gave demonstrations and distributed CDROMs at both ADASS in Hawaii and at the AAS in Atlanta. At the ADASS, we had eight AIPS++-related posters, and Darrell Schiebel gave an invited talk on ``Event-driven programming with Glish''. An interesting point that was apparent from various ADASS papers, was that basic design of AIPS++ as compiled language plus an interpreted glue language is now being adopted by other groups.
As described above in the Single Dish section, Joe McMullin visited Arecibo in February 2000 to give a talk and demonstration of AIPS++, and in particular the dish program. This appears to have worked out very well, and we now have a core group of enthusiastic and very helpful users at Arecibo.
There are some straightforward lessons from from our outreach activities:
Considerable effort has been devoted to improving the installation process, particularly in the area of robustness. For example, Darrell Schiebel improved the way shared libraries are handled to avoid confusion with shared libraries already present.
We have established how the new features of the C++ standard will be used in AIPS++. A particularly important area is the improved handling of casts.
The global makedefs file has been improved by putting commonly used commands from the local makedefs in it. For example, linking with purify (and purecov and quantify) has been incorporated.
In Socorro a few licenses of Insure++ have been bought as an addition to the Rational products purify and purecoverage. It looks as if it does test more thoroughly. Unlike purify it inserts test code at compile-time, which makes it compiles to test only parts of the system. The Rational products work fine with egcs 1.0.3, but not with 1.1.2. Insure++ works for both compilers on all platforms.
There were two personnel changes: Joe McMullin was named chief tester and Bob Garwood was named code cop. Bob replaced Ralph Marson who had performed the duties of code cop with enthusiasm and energy.
During the last cycle, note 228 was written. This note describes our testing mission and implementation along with future testing plans. A set of scripts was written which check the health of various installations daily and weekly so that regular reports can be sent to aips2-lib. A new version of note 167 and the c++ review rules were checked in to the system. All occurances of the old arraycast were removed from the system. The easy (name *) 0 casts were changed to the new c++ casts as an example for users. The code review of the first part of the Lattice module, the Coordinates module, the DisplayOption class, and the TableMeasures module were largely completed during the last cycle although the actual movement of this reviewed code out of trial will happen early in the next cycle.
Reviewing of several classes was completed. Most importantly, the reviews of the TableMeasures module and much of the Lattice module were completed. Regular monitoring of the unit testing and glish assay tests by McMullin continues.
David Barnes and Harold Ravlin made substantial improvements to infrastructure and applications. TrueColor X Windows displays are now fully supported, and multi-color-component output capability has been added to the PostScript software. These additions made possible the provision of multi-color-component overlays in the viewer tool - such overlays can be used to produce stunning images from multi-waveband data. As well, support for image masks was added at all levels, so masked image data can now be easily displayed by users. An early version of a DisplayData class for plotting catalogue information was provided, and the viewer controls were reworked and improved, especially those for "animating" multi-dimensional images.
The first release (version 1.2) was made October 4, 1999 at the ADASS meeting in Hawaii. Since there we have distributed about 1400 CDROMs split roughly equally between Solaris, and Linux 2.0, and Linux 2.2. The total number of defect reports submitted (since May 1999) now approaches 1300, of which we have fixed about 1100. The remainder are postponed until the development cycle that is now starting. In the version 1.3, there were no unresolved important defects (severity level 1 or 2). See Appendix G for a report on defect statistics.
It is very hard to estimate the number of active users of AIPS++. Our request that users register resulted in only 25 new registrations, with many known users missing. Looking at the defect submissions, we find 29 non-developers plus an effectively anonymous category of probably about 10 more people. Thus the total number of external people who have used the package enough to submit a defect report is probably about 40. This means that we are still in the ``early-adopter'' phase where only a small fraction of the final audience is using the package. This is reflected in the usage at consortium sites, which is still quite limited. Based upon a variety of sources, we expect this to pick up once main-line synthesis calibration and imaging is completely present.
The second release (version 1.3) is planned for mid-April 2000. At the time of writing, it is on schedule. Preparation has been much simpler this time around since most of the thorny issues seem to be have been dealt with during the first release. We plan to support Solaris and Linux only, again, and to duplicate 1000 CDROMs of each for distribution via various mechanisms.
The major changes in release 1.3 compared to 1.2 are:
| Package and Module | What |
| general.deconvolver | MEM and multiscale deconvolutions |
| images.imagefitter | Interactive image fitting |
| images.imagepol | Polarization calculations |
| measures.comet | Handling of JPL comet ephemerides |
| utility.fitting | Linear and non-linear fitting |
| display.viewer | Sundry viewing capabilities |
| display.plotter | PGPLOT-based plotting |
| synthesis.componentmodels | Construction and use of discrete components |
| synthesis.flagger | Simple flagging of visibility data |
| synthesis.imager | MEM, Multiscale deconvolution |
| synthesis.imager | Mosaicing, and Single Dish/Int combination |
| synthesis.calibrater | Initial synthesis calibration |
| synthesis.dragon | VLA wide-field imaging |
| synthesis.simulator | Simulation of visibility data including errors |
| dish.dish | Single dish reduction program |
| nrao.vla | VLA filler |
Planning for the third release (1.4) will occur at a face-to-face meeting of the lead developers in late April. The main goals will be to continue the integration effort, to improve substantially the synthesis calibration and editing capabilities, and to deploy some new visualization capabilities.
We also plan for mid 2000 a developer pre-release to be based around Linux only. This will serve two purposes: to give ``early-adopter'' developers a way to work, and to develop the necessary mechanisms needed for the full developer release. The main work required for the pre-release is to improve programmer documentation, and to check the build parameters against standard versions of Linux (probably RedHat 6.*). Based upon the number of requests for a development system, we expect that perhaps 10-20 people would use the developer's prerelease. It is not our intention to support an open-source development model.
In late 1998, we were approached by the ACSIS development group at HIA/DRAO regarding the use of AIPS++ in their development of reduction software for the JCMT ACSIS correlator. They were investigating the use of AIPS++ for the correlator and wished to know if we would be able to provide some support. Discussions continued throughout the first part of 1999, and in August 1999, we agreed on a set of tasks that would be performed by AIPS++ to support the HIA development. Since the ASCIS group was very experienced in C++ and the use of Glish, the support only amounted to about 2 months of Darrell Schiebel's time. Our motivation had two parts: first, the ACSIS development would provide a good test bed for many aspects of AIPS++, and second, the ACSIS system would demonstrate to the community the capabilities of the AIPS++ toolkit. The development would also be part of a natural progression in that the final system would be close to the Parkes Multibeam system in capability but would be able to handle a data rate about 50 times higher.
The chief software architect was Tony Willis, who writes: ``For the past 16 months or so, we have been using the AIPS++ toolkit to develop a large fraction of the reduction software for the JCMT ACSIS correlator. ACSIS (Auto-Correlation Spectral Imaging System) will be capable of computing the spectra over 1 GHz bandwidths for up to 32 receiver beams every 50 millisec. The goal of the reduction system is to produce calibrated, gridded 3-D data cubes that can be viewed in real-time or near real-time.''
``I believe that our use of the AIPS++ toolkit has been beneficial to both the ACSIS project and to the AIPS++ group itself.''
``The ACSIS reduction system makes heavy use of distributed processing and message passing between tasks running on different computers. We require something like 40 tasks (c++ glish clients) to be running and communicating with each other at the same time. Also, we require that message passing between clients be dynamically configurable at run time. It has proved easy to meet these requirements with the AIPS++ 'glish' message-passing system. This message-passing system, as well as the glish-Tk GUI package, has proved extremely easy to set up and use. The AIPS++ toolkit provides an extensive c++ library of classes for astronomical and radio-astronomical applications. The existence of this class library, together with glish, has been of enormous help to ACSIS programmer productivity.''
``The AIPS++ group has been very supportive of this project. Early in the project we identified a number of areas where the AIPS++ toolkit was deficient, or had bugs, that affected the ACSIS system design. The AIPS++ group provided fixes for these problems by an agreed-upon deadline, so that ACSIS development has been able to proceed on schedule.''
``Because ACSIS is the first project to actually use a glish-based distributed processing environment spread over many computers, our testing revealed several glish bugs/defects that would not have been apparent to the regular AIPS++ developers. They have focused on development of an integrated application running on a single workstation. Our discovery and reporting of these problems will make the development of the next distributed system based on AIPS++ much easier.''
In this cycle we had 4 people working in AIPS++. These are (nominal time allocations given) Neil Killeen (75%, also local manager), Wim Brouw (75%), Mark Wieringa (30%), and David Barnes (75%). Wim's work in ATOMS has now ramped down to something modest (some testing left to do) so his AIPS++ availability continues to ramp up again.
Following the public release, we installed a range of systems throughout the ATNF. We have the base release, the stable release, the development release (weekly and daily varieties).
We don't have an egcs103a installation (required for Purify products). We rely on the Socorro installation for this. The link is just tolerable for basic displays. Unfortunately we are unable to use Purify on the Display Library shared object (this is the main reason for not making an installation here). This is a significant problem.
David's time was spent on
Wim's time (32% in Q4 1999 and 54% in Q1 2000) was spent on
Neil's time (75%) went on:
Mark's time went on
See:
http://wwwnar.atnf.csiro.au/
wiering/skasim
During this cycle at NCSA and BIMA had two individuals working on AIPS++. Doug Roberts (the local manager) contributed 85% and Harold Ravlin contributed 65% to the AIPS++ project.
The computer systems used at NCSA include Linux, Solaris, and SGI machines. The SGI Onyx, which is mainly used for visualization has had a few problems with a couple of disk drives and the graphics board went out in November; since then it has been stable. Harold Ravlin has worked on backup scripts for all the NCSA/AIPS++ machines in the Astronomy Department. The SGI Origin systems at NCSA have been stable. We have three builds on the Origins which insure that we always have two stable systems. Now that Wes Young has enabled the 64-bit compilation using the native compiler, we have a stable 32 and 64 bit system as well as a test build system.
Harold Ravlin has spent some time during this cycle to deal with our system problems. His AIPS++ work has been to expand the AIPS++ visualization facilities to to deal with TrueColor displays.
Doug Roberts has spent some time away from the parallel project by working on the BIMA imaging pipeline: design and identifying AIPS++ requirements. The rest of the time has been spent working on parallelization targets.
Barry Maguire writes:
``I began work at Jodrell Bank in early February 2000, primarily to develop AIPS++ software for the reduction of MERLIN data. The first step was to make a trip to Dwingeloo to find out what AIPS++ development their people were doing for VLBI and make sure we were not working in the same areas. As you know, at the end of February, Jodrell Bank rejoined the AIPS++ Consortium and I set up a consortium installation at the JBO site. It was necessary to repeat this process a couple of times as the required dependant software libraries were not already in place. But after a couple of attempts and a fair bit of help from Darrell Schiebel I managed to get the installation running successfully. The only thing missing from the installation is the automount to the Master repositories, as our system administrator expressed some concerns regarding the University firewall, but we hope to have that mounted in the next few weeks. The areas of interest for MERLIN are in data calibration, editing and UV-visualization, but of course a neccessary first step is to ensure that MERLIN data can be read into AIPS++, and so the first task is to build a MERLIN Filler. Having spent a few weeks trying to familiarize myself with the system and programming manuals and particularly focusing on the MeasurementSet and associated filler classes, I have now made a trip to Socorro for a couple of weeks. I have found my stay here very instructive as I now have a good feel for how AIPS++ code is implemented and how the class hierarchy is structured, with the highest level interface being built into a distributed object and bound to Glish, thus allowing the users control via the command line or GUI interface. This is something I picked up in a few days here at Socorro, but would have taken me a lot longer to grasp staring at manuals back at Jodrell. While here I have begun work on a MERLIN MS filler, basing the structure on the VLA filler, although I am somewhat hampered by the fact that the developers at Jodrell are currently developing new on-line software for the MERLIN correlator that is due to be installed this summer. When it is, the output data format will be markedly different from that which currently exists and so the format of the data to be read by any filler is presently uncertain. I can write a filler based on the existing format now, but it will require substantial changes later this year when the new software goes on-line.''
Local project members: Ger van Diepen (GVD), Jan Noordam (JEN, local manager). Associated member: Tom Oosterloo (TAO)
This cycle, the NFRA contribution in FTE's was as follows:
| Global AIPS++ | Local AIPS++ | Other | |
| Ger van Diepen | 25% | 45% | 30% |
| Jan Noordam: | 10% | 60% | 30% |
| Oleg Smirnov: | 60% | 10% | 30% |
(NB: The 'other' category includes vacations)
Oleg Smirnov has arrived and has started work on ionosphere calibration. In general, he will help to 'widen the thin path' through the uv-data processing part of aips++.
Tom Oosterloo has further developed his ideas for a 'calibration pipeline', which will be implemented as a joint project with ATNF. Its modules should be directly usable in the global package.
In this cycle, the most substantial contribution to the global package was again made by Ger van Diepen in his role as Technical Leader, and as the man responsible for a number of essential infrastructure modules (tables, lattices, ms2uvfits, etc). This is eleborated elsewhere in this report.
In the meantime, the upgraded WSRT continues to rely heavily on AIPS++. The online system (TMS) uses AIPS++ Tables, Measures and Glish, and MS2uvfits for data export. The local WSRT 'mini-package', written in Glish by Jan Noordam, is operational and widely used locally for inspection of uv-data and determination of instrumental parameters prior to observations.
The TableMeasures classes are now actively used in the MS v2 and in the ComponentModels. It appeared that a few small changes were needed to make the framework fully suitable for the MS classes.
For the ACSIS people (in particular for John Lightfoot) the Table Query Language has been extended with the possibility to operate on TableRecords. It makes it possible to use a TaQL-command to select TableRecords from a set of TableRecords.
The Table system has been extended with a NoReadLocking option. It makes it possible to read a file without locking it. Arthur Coolen has actively worked on a new storage manager which will be more robust and use less memory. It is expected to be ready in May 2000. A widget has been developed to help the user in forming a TaQL-command. It makes it much easier to select records from a table.
The logger has been extended considerably. The user has much more control how the widget looks like. The main change is the addition of a query possibility. It uses the TaQL-widget to let the user form the query string.
Michel Caillat in Meudon and Christian Bruens in Bonn would like to develop code in AIPS++. Various email messages have been exchanged to solve problems in setting up the system and to tell them about the AIPS++ software. In May Michel will visit Dwingeloo.
The core NRAO group is now Braatz (75%), Cornwell (75%), Garwood (75%), Golap (75%), Holdaway(75%), Kemball(75%), Marson (75%), McMullin (75%), Minter(75%), Schiebel (100%), and Young (100%). Golap and Young are funded by NCSA-NRAO collaboration. Minter by a three year NSF visualization grant. We have one open position funded by the NSF grant for visualization. George Moellenbrock will join the AOC group in late May to work on synthesis development, replacing Peter Barnes who left at the end of 1999.
Jim Braatz' primary responsibilities are to manage the Green Bank installation of AIPS++, to support GBT operations, and to work on single-dish applications. Specifically, Jim worked on the following tasks during this cycle:
I spent a larger amount of time this cycle on management and less on actual development.
In management:
My coding contributions have been some fraction (about 30%) of the targets that I undertook at the beginning of the development cycle.
Bob Garwood's primary responsibility is to oversee and contribute towards the single dish work in AIPS++. This work remains focussed on the DISH environment and the support of the GBT. His contribution in support of the GBT is primarily through the GBT fillers which convert the GBT FITS data files to an AIPS++ MeasurementSet. In addition to these duties, he is responsible for the maintenance and enhancement of the FITS classes and, as author and maintainer of the GUI guidelines, he spends some amount of time reminding other AIPS++ workers on the importance of following those guidelines. He is also the chief code cop and head of the QAG.
Over the last development cycle, he has done the following:
Mark Holdaway has made a number of modifications to the maximum entropy code, including the addition of masks and correct image plane operation for single dish deconvolution. In addition, he cast the MEM algorithm into a form which reflects AIPS++'s incremental model of deconvolution, but calculates the entropy function on the cumulative image. This last enhancement is required to get MEM to work for the wide-field and multi-field contexts.
Several problems in the SkyEquation which limited the quality of mosaiced images were found and fixed.
Now most of the deconvolution algorithms supported by AIPS++ are enabled for the multi-field context. Mosaicing can be performed using Clark, Hogbom, or multiscale CLEAN, maximum entropy, and maximum emptiness. Minimum-sized FFT's in mosaicing have been succesfully prototyped.
Feathering has been added to imager, permitting the addition of total power data to mosaiced interferometric images.
Mark has done work on the simulator, including adding an image evaluation tool to compare the initial model image with the image constructed from the simulated data.
Mosaicing simulations for ALMA will require time dependent, antenna dependent pointing errors, which result in a time and baseline dependent primary beam, and a different Fourier transform for each visibility. Most of this code has been written, and this work was presented at the ALMA Configuration Conceptual Design Review in Tucson.
Also using the simulator and imager, Mark has demonstrated the construction of time dependent voltage patterns generated from phased arrays such as the SKA may utilize. The phased array voltage patterns would then be used by the simulator to predict the actual data. An outline of this work, plus how imager would generate accurate images with time dependent (and to some extent, unknown) voltage patterns, was presented at the SKA US Consortium meeting in Aricebo.
I have continued coordination of synthesis and parallelization in this development cycle, as described in the separate sections covering these areas in this report. I have also continued with coding projects to meet individual development targets to which I am assigned, or for which I am jointly responsible. These include synthesis integration above calibrater and imager as part of the effort to expand automated and pipeline reduction within AIPS++, finalization of the MS v2.0 transition, work on parallelization infrastructure in C++ required for wide-field parallelization, and general testing, defect identification and correction. I have continued work in the system area in specific areas, including the testing of Rational Purify, Quantify and PureCoverage release versions with revisions of the egcs and gcc C++ compilers, evaluation of other memory leak tools, including Parasoft Insure++, supervision of the trial port to Windows NT, maintenance of two Solaris builds at the AOC as local site manager, and other system administration, computer procurement and general administration. I spent some time this cycle arranging the meeting, "Imaging pipelines in radio astronomy" (Jan 18-20, 2000), in follow-up to this meeting and in user support and tester meetings. I participated in a visit to several sites in Europe with T. Cornwell in November, 1999, as part of our continuing public outreach initiative. As customary, assistance with release testing and preparation has taken a signficant fraction of time towards the end of the development cycle.
Ralph Marson's aips++ responsibilities continue to be the development of radio interferometry applications. For the month of January he was the Linux site-manager in Socorro.
During the last three months the main application Ralph has worked on has been the vlafiller. Work completed includes:
The other major application area Ralph has worked on in the last three months is the ComponentModels module. This includes the following work.
All the test programs I am responsible for in the trial package were overhauled so that that they now run as well as they ever did.
A number of bugs were addressed in the above mentioned areas (vlafiller and components) and in the mathematics module.
four weeks of science time was taken in this quarter. Science work included:
Visited the observatory and staff. They had constructed a list of wish/needed items for their integration of AIPS++ into the Arecibo Observatory. Developed a response to the various items including a development plan for the unfinished items. Also: 1) delivered a proto-type RFI tool; 2) Gave a talk on AIPS++/DISH to the scientific/computing staff; 3) Gave a talk on astronomical programming tactics in Glish for the same; 4) installed a development version of AIPS++ with regular inhales (it took Bob's efforts to really make this happen); 5) Produced a document for them on tool building and an example Arecibo tool as a template for layering their analysis needs onto and using DISH; 6)Provided examples of plugins and scripts for extending the functionality. We also established regular conference calls to maintain close contact/collaboration over the next development cycle.
Darrell's primary responsibilities include maintenance and development of the Glish scripting language, system level release preparation, and about half the work (shared with G. van Diepen) of maintaining the code distribution and version control system. Each of these received some attention during the cycle:
In this section, we give the names of people in the various AIPS++ groups and the nominal fraction of time allocated to AIPS++.
| Global AIPS++ | Local AIPS++ | Other | |
| Ger van Diepen | 25% | 45% | 30% |
| Jan Noordam: | 10% | 60% | 30% |
| Oleg Smirnov: | 60% | 10% | 30% |
Thus, in aggregate, we have 21 people contributing about 14.45 FTEs to the AIPS++ Project.
Here we show the defect arrival and repair rates from May 1999 when we adopted the ClearDDTS program (replacing the GNATS program).
