 News |
FAQ
|
| Search
|
Home
|
| Getting Started | Documentation | Glish | Learn More | Programming | Contact Us |
|
| Version 1.9 Build 1367 |
|
T.J. Cornwell, NRAO
1998 July 17
In this report, we cover the personnel changes and progress in AIPS++ over the last quarter (1998 Q2). We also discuss the issue of ``Thin Path'' development raised by the STAG and mentioned in the previous quarterly report.
http://www.nmt.edu/~pio/dan/dan-play.html.
As a result of all of these changes, the next quarter will require lot of work on hiring. In addition to replacing Dan Briggs and Jeff Uphoff, we have to hire a post-doc to work on wide-field imaging, and someone to work on parallel i/o for the parallelization effort.
The loss of Dan Briggs will slow down our work on wide-field imaging, and the departure of Jeff Uphoff will further freeze our work on system support.
In Single Dish support, we added the following to the Dish program: an operation to save results back to disk, the ability to save the state of Dish and its plotter between executions of the program, and use of the same pgplotter widget in the dish plotter as used elsewhere in Aips++. The full SDFITS convention is now supported in the fits2table application. At Green Bank, we added support for the Holography backend to the existing fillers. The RFI data base developed last quarter is now in use in Green Bank and we are responding to the initial user feedback. We support GBT staff in on-going testing of GBT hardware and software. We have begun development of the continuum analysis tools necessary for GBT commissioning.
In Synthesis support, we finalized the definition of the second version of the MeasurementSet format. This will be implemented following the third beta release. We developed a glish-based visibility plotter using the ms module and the pgplotter interactive plotter. We expect that this will meet visibility examination needs for the short term but will have to be revised or replaced when the display library is available for application programmers. Only about a weeks investment of time was needed for the development of a very functional initial version. The sky synthesis imaging application was subject to testing by various testing and as a result a number of (mostly subtle) bugs were fixed. The corresponding cal synthesis calibration application is still under development and has not been placed in the system. Wim Brouw developed a very promising approach to wide-field imaging in which some of the most time-consuming operations (the image coordinate system conversions) are no longer required. The Measures system now supports this via the UVWMachine class, and we expect to implement the full algorithm in the next quarter.
In Glish, a number of important modifications were made:
There were several reasons for making this change:
This new client has been checked into the AIPS++ system, and after a long period of testing, the Tk widgets will no longer be linked into the Glish interpreter.
In Measures, after ample discussions about the proper inheritance structure and format, we wrote (and got code-reviewed) QuantumHolder and MeasureHolder classes, which are generic holders of any Measure or Quantum, and aimed at providing a general interface to the Tasking system. We included the VLBI reference source list in the 'data' system, which is accessable from both the Glish (measures gui) and c++ code. We wrote the MBaseline Measure (and its coordinating classes): no Earth tide or other secondary effects included as yet, Muvw Measure (and its coordinating classes), added an angular shift method (in both true and projected angles) to the MDirection Measure, added de-projection of an 'off-axis' field to UVWMachine, improved some of MDirections forward/backward conversion precisions, finished EarthMagnetic field Measure using the IGRF reference magnetic field, wrote a FieldMachine to easily calculate the magnetic field at a certain direction in e.g. the ionosphere, wrote a VelocityMachine to convert between velocities and frequencies in a simple manner, and added new ITRF and TOPO codes to MDirection. We also wrote an article for newsletter on measures gui. In AIPS++ Infrastructure, mostly minor changes were made, mostly in response to bugs. The tablebrowser was improved significantly, adding user-controlled formatting, smart formatting of special column types such as time and direction, etc.. The catalog browser was substantially re-written in response to comments and bug reports from the alpha testers. Finally, we developed a prototype of a command-line parameter-setting shell, similar in principle and details to various systems now in use in astronomy (Standard commands package, AIPS, MIRIAD).
In FITS support, Harro Verkouter of JIVE continued his work on re-implementation of some of the key FITS classes. This is needed prior to the writing of an improved UVFITS reader and writer.
In Image Analysis, work has largely revolved around regions of interest, masks, the classes which hide the details of optimal iteration through images, and the integration of these things with our high level applications. Neil Killeen and Ger van Diepen made good progress in these directions, and now have in place:
In Display work, David Barnes has joined the ATNF aips++ team and reinvigorated the Display Library (DL) work following Tom Oosterloo's departure last year; he is working closely with John Pixton of NCSA. David spent time familiarizing himself with the DL design and current implementation. Then, he
In the System area, the major change this quarter was the addition of a system to "bless" development updates as having met internally-defined criteria for stability and usability; this now enables AIPS++ sites that require stable installations to confidently track development progress between public releases. Other system changes include: continued incremental improvements to the code management, distribution, compilation, testing, and documentation systems; improved support for providing some of the third-party software packages required by AIPS++ to end-users, particularly those running Linux; and an investigation into building AIPS++ with the free Lesstif library suite, attempting to eliminate its dependence on the commercial Motif library suite.
Some future plans include: investigating replacing the current code management and distribution systems with a free, "off-the-shelf" alternative such as the widely-used CVS package; continued incremental improvements to project "infrastructure" components; elimination of AIPS++'s reliance on Motif; and simplified installation procedures for AIPS++ on the major supported architectures.
In the area of Parallelization, work proceeded on a number of fronts. In April, Wes Young completed the parallelization of the spectral line Clark CLEAN application in AIPS++. Using the spectral line deconvolution Doug Roberts has spent much of April, benchmarking the task to measure the speedup across processors on the NCSA SGI Origin 2000. The speedup achieved is encouraging; the application executed 22 times faster than uniprocessor speed on 32 processors. An ideal speedup on 32 processors would be 32. The results of the benchmarking were presented by Doug Roberts at the Alliance98 conference, held in Urbana at the end of April. Brian Glendenning and Ruth Milner from NRAO along with Dan Briggs, Dick Crutcher, and Doug Roberts from NCSA attended the Alliance'98 conference. The on-line version of the Alliance'98 presentation is available at:
http://pacont.ncsa.uiuc.edu/A98/poster/roberts/.
Several potential connections were initiated at the conference. An NCSA team which has developed tools to analyze I/O usage is interested in helping to instrument our code. NRAO is looking to hire someone for this in Socorro who would collaborate closely with the NCSA team to identify I/O problems. The parallel-I/O component of the MPI-2 standard is available now and will be investigated by the new hire to determine if it can help the I/O performance of AIPS++.
In Documentation, we continued with publication of the AIPS++ Newsletters:
http://aips2.nrao.edu/aips++/docs/html/navpages/learnmore/newsletters.html.
Publication of a new edition of the newsletter is now announced via a mailing-list (aips2news@nrao.edu). We have installed a new system of web pages that have improved layout and navigation aides. Our alpha testers (particularly Rupen and Anantharamaiah) have started re-writing the introductory documentation, with a view to making the concepts in the system comprehensible to astronomers, and to defining terminology to be used elsewhere in the documentation.
In Management, Brian Glendenning left his position as Deputy Project Manager to join the NRAO MMA project. Athol Kemball will replace Brian as Deputy Project Manager, effective July 1. Ger van Diepen will replace Brian as Acting Technical Leader, effective September 1. Tim Cornwell and Brian Glendenning gave presentations at the NRAO Visiting and Users' Committee and Tim Cornwell gave a presentation to the AUI Board of Trustees. We also performed an analysis of the our success at keeping target dates. During the last three years, we have identified and met 433 targets. The quartiles are 75% are late by 1 week or less, 50% are late by two weeks or less, and 25% are late by 5 weeks or more. The tail of late dates ranges up to a few hundred days but these are mainly targets that have been deferred following a change of plans. There is no systematic trend in target date lateness with absolute time (i.e. we have not got worse at meeting targets). Since the average target takes 4-6 weeks to finish, the fractional delay is about 33-50%. This noise will clearly make planning on longer time scales problematical. The cause of lateness is hard to identify but our belief is that much of the delays are due to time being taken up by other, unscheduled activities such as bug-fixing and making unforeseen changes to nominally completed parts of the system. One might then regard the lateness as the cost of complexity. As a result of this analysis, we are now tracking time spent on- and off-target, with a view to improving our overall performance in meeting targets, and thus easing planning within the project. In the future, we will collect statistics on times for fixing bug-reports, as this is more important as a quality measure from the point of view of users.
Preparation of the third beta release continues. The pacing items now are the release of the cal synthsis calibration application, the continuing testing of all modules, and the documentation being written by our testers (particularly Michael Rupen and Anantharamaiah). The planned documentation consists of a rewrite of the introductory material and the establishment of astronomer-preferred nomenclature. Retrofitting of the new terminology to the existing extensive documentation will also be needed.
Our expectation for the third beta release is that the vast improvements in functionality and interface over the previous (second) release will attract comments from both new users and our existing beta testers. We believe that continued tuning of the user interfaces will be the most important issue arising from the release, and with that in mind, we plan to re-allocate some fraction of Darrell Schiebel's time to replace Brian Glendenning as our key worker on user interfaces.
We expect that at least one more beta release will required following this third release, putting the probable time for the first public release some time early in 1999.
This section discusses the question of so-called "thin-path" development in AIPS++, and the implications for current priorities and resources. This topic was raised by the STAG in the recent meeting, and discussed extensively both within the project and with various advisory bodies such as the NRAO Visiting and Users' Committees. Here we give our final statement on our approach to thin-path development.
At the current stage of development of the project, and for a planning interval of 6-12 months, the following key synthesis development priorities are identified:
Significant synthesis capabilities exist within the package at present, but are not sufficiently widely used in the scientific community. We identify the following reasons:
We note that significant progress has been made on many of these items over the course of this year, and the primary focus of the project at this time is application development for scientific use. Thus the balance has been shifted away from infrastructure development, where prudent and possible.
The internal synthesis interfaces, which form an important part of the synthesis design as a whole, need to be finalized for the core synthesis capabilities, so that more advanced applications can be developed by a broader range of developers. Such applications include three-dimensional imaging and mosaicing, as examples. The interfaces are more efficiently tested by developing advanced prototype applications of the type considered, by evaluating the interface design on this basis, and also by ongoing general design review and testing.
It is important that the project meet time-critical targets required for successful AIPS++ use at consortia sites, particularly where the use of AIPS++ is closely integrated into the critical path required for instrument use or operation.
An overall synthesis development strategy has been adopted in which the key strengths of the AIPS++ system are used to achieve the priorities listed above. These strengths include the programmability and flexibility offered by the command line interpreter (Glish), and the system architecture overall, as well as the fully general calibration and imaging model. AIPS++ offers both improved user and system programming as the command line, and greater communication between the command line environment and applications.
The facility for greater programmability can be used to package the existing synthesis capabilities in a manner which allows greater scientific use. This includes the production of scripts to automate basic calibration and imaging. Further user interface improvements can and have been achieved through a customized graphical user interface for important synthesis components (e.g. sky), and the implementation of a command line parameter-setting shell. A separate requirement for greater access to the package is that of targeted documentation for scientific users.
The general calibration and imaging model allows new and unique scientific capabilities to be implemented as a result of the more sophisticated formalism. The strategy for adding new capabilities can take two directions, namely:
In the second option, data are transferred to the new package at a well-defined point, for further reduction, with this point moving, over time, to cover a larger fraction of the existing packages. In this scheme, new scientific capabilities not already available in other systems appear earlier, and have higher precedence, than in option (i). Most data reduction packages for radio-interferometry developed to date have followed development path (ii), and this has been the general AIPS++ synthesis strategy.
The question considered here is the change required to the overall synthesis development plan to accommodate a thin-path capability for connected-element data as a short-term priority. The components and resources required are summarized next.
The primary changes in the current synthesis development plan to implement a thin-path capability lie in the following areas:
Calibration and cross-calibration have been the subject of active AIPS++ development this year and are consequently not considered as part of the thin-path components. The core components of these capabilities form part of the planned beta-3 release. Imaging and self-calibration has been available for some time in AIPS++, and are similarly not included.
Thus, the thin-path capability requires a change in the priority of the elements required in early data reduction. Primary amongst these would be the implementation of a VLA filler. Note that other connected-element instruments including WSRT and the ATCA already have preliminary fillers, written by local consortia members. In some cases these are simpler to implement, due to a smaller number of archive formats or related restrictions. Experience at NRAO has shown that data fillers in other packages supporting a full range of archive formats, and multi-volume input media have been a significant effort for the VLA and VLBA. This is mostly related to the evolution of the on-line output formats and the mismatch between the on-line and off-line data format requirements. In addition, concatenation of multiple tape volumes to existing files on disk adds further complexity, as information is revealed to the filler on a per-record basis and not for the dataset as a whole.
Data display utilities include those required for all the customary graphical displays of uv-data (eg. visibility time series, spectra, closure data), and the capability to list data in tabular form. Interactive editors are those applications which allow uv-data to be edited in a fully-featured, interactive graphical environment including data edited per baseline in line-graphic form, or stacked in axes of time, frequency or baseline in a raster TV display.
The associated utilities listed above, refer to those applications required in support of early calibration or editing. Examples include utilities to set and determine absolute flux density scales, or correct for antenna or source position errors.
The definition of what constitutes a thin-path is problematic in practice. It is highly dependent of the nature of the observations being processed, and may differ significantly even within a single category of observations (eg. continuum). It requires the basic capabilities listed above. If the thin-path capabilities are inadequate there is the real risk of alienating users or pulling the project in directions away from fundamental work which needs to be completed in imaging and calibration. This would limit the overall use of the package if neither option received sufficient resources to achieve critical objectives as a result. The implementation of makeshift thin-path capabilities would ultimately prove expensive, as they would need to be rewritten in the future.
A hybrid development plan to address the thin-path question is described next.
The project is opposed to a full-scale diversion of effort into early processing capabilities at this stage, due to the risk this poses to fundamental work underway in calibration and imaging, and consequently the overall scientific use of the package. However, an approach has been adopted and is described here which meets key thin-path requirements without compromising broader development.
The key elements of this strategy include:
This plan can be achieved as a result of the recent expansion in the number of synthesis application programmers in the project, consistent with the increased emphasis of applications over infrastructure, and some diversion of effort from VLBI priorities. The new synthesis programmers are Mark Holdaway and Peter Barnes. It is unclear on what timescale Dan Briggs can be replaced.
In this approach, Ralph Marson (25%), assisted by Athol Kemball, will start immediate work on a basic VLA filler for the limited public release. The design will be approached such that components can be re-used for a VLBA filler at a later time. The development of basic data display modules consistent with the image display library requirements, allows work to proceed on automated data reduction scripts and simple interactive editors using the same display components. The interactive editors will not offer all features available in other packages but should meet the thin-path requirement. Most importantly this approach allows the goal of scripted processing to proceed unimpeded, and to offer a new, attractive scientific capabilities which are part of the current synthesis goals listed earlier. Peter Barnes will assist in the area of display utilities.
Utilities associated with early processing, such as flux density scaling and baseline corrections, will be scheduled at a higher priority within calibration development.
High level work in calibration (A. Kemball/M. Wieringa/W. Brouw) and imaging (3D-D.Briggs' replacement; Mosaicing-M.Holdaway/W.Brouw; General imaging and solvers-T.Cornwell) will proceed essentially as originally planned, thus preserving the original goals in synthesis development.
The primary cost in this approach is the slower development of more advanced calibration capabilities, especially in VLBI. Note however, that VLBI development will not be terminated in its fundamental form (such as fringe-fitting). Work on more advanced capabilities such as phase-referencing or geodesy will have to be deferred however, as will more advanced calibration solvers in connected-element reduction.
In Single Dish, the Dish program will be included in the next beta release. This will fully support single dish FITS data into Dish. Existing Dish operations will be made to work on multi-polarization data. It is expected that both Parkes and Arecibo staff will being testing this version of Dish and providing the first real-user feedback of Dish. An initial set of GBT commissioning tools for continuum data will be ready by the end of July. The internals of Dish will be rewritten to make it easier to use Dish through the command line i nterface as well as to make it easier to add operations and site-specific features. The existing gbt fillers will be reworked to improved their speed and usefulness with on-line data. Work will start late in the quarter on Single Dish imaging using the same imaging scheme as used in Sky.
In Synthesis, we will continue development along the lines spelt out in the synthesis development plan, and described above in the section on Thin Path development.
In Glish support, In the next quarter, we hope to perform the following, relatively straightforward modifications:
In Image Analysis, the major task is to cement the region and mask handling infrastructure and propagate it to Glish with a good interface for users. We plan to:
In the Display Library, the next quarter should see the DL evolve into a "very-close-to-useable" aips++ module, with capabilities of displaying rasters and contours of 2-dimensional slices of n-dimensional Lattices taken along any of the pixel axes. Support will be available with such displays for colour table manipulation, zooming, and cursor feedback. We will also develop a schedule for the generation of applications thereafter.
David Barnes will visit John Pixton at NCSA during this quarter. Unfortunately, John leaves the NCSA to work in Norway in early August. David's visit is partly aimed at maximising their coordinated output, but also to maximize knowledge transferrence to David. It is hoped that DL infrastructure will be largely in place by the time John leaves, but no doubt there will be some loose ends. John's considerable talents will be missed. In Infrastructure, Darrell Schiebel will take over Brian Glendenning's work on the user interface, adding a number of facilities to the GUI and also developing special purpose widgets, e.g for inputting Measures such as Directions.
In Measures, we will continue to improve the implementation in various ways and add some new features as needed for applications:
Wim (our sole contributor to Measures) will have to spend some time on SOFA work in the next quarter. Up till now it has mainly been limited to comments on work by others, but some coding and/or testing is imminent.
This quarter the ATNF had 4 people working in AIPS++. These are Neil Killeen (25-75%, also local manager), Wim Brouw (75%), Mark Wieringa (30%) and now David Barnes (75%) as of April 6 1998.
We welcome David to the ATNF and aips++. He is stepping admirably into the role of responsibility for the Display Library and its applications. The DL has been largely dormant since Tom Oosterloo's departure last year. Note that Wim's nominal contribution of 90% has been decreased to 75% owing to growing 1KT committments.
Our systems (Solaris/Gnu and Digital Unix/Gnu) were generally stable this quarter and required close to zero maintenance. However, the documentation system became unstable towards the end of the quarter (for all consortium sites), and we have been several weeks missing the users reference manual, the glish manual and the notes and memos series. The problems arose through the non-trivial interaction of TeX, LaTeX, dvips and latex2html.
NCSA now appears to be making regular native IRIX (SGI) inhales. We can probably now consider another attempt to build on our SGI machine.
Athol Kemball visited us for three weeks. He spent two weeks in Narrabri working with Mark Wieringa, and one week in Epping mainly working with Wim, but also with David and Neil. We all found Athol's visit very rewarding.
Neil Killeen spent two days in Dwingeloo working with Ger van Diepen on issues to do with regions of interest in lattices.
There has been no significant new system testing this quarter at the ATNF (of course the multibeam people still process their data with their Glish scripts built on aips++, and Glish/Tk forms the basis of the Parkes Telescope Control System user interface). We do have three users who have expressed strong interest in using the single dish part of aips++ to analyse their data (ATNF single dish software is not overly impressive). We are awaiting the final connections in "dish" to handle reading and writing Single Dish FITS files, and multiple polarization handling. This work is expected complete by mid July 1998.
Wim Brouw has spent his time this quarter on:
Measures:
Various:
Bugs and problems:
David's time was spent on
Neil spent one month away in Bologna having some science time. Neil's aips++ time was spent on:
Mark had little time this quarter for aips++ as a result of Gamma ray bursters, work on local Year 2000 issues and other local work. He has spent his aips++ time on:
During the past quarter we had 5 people (2.8 FTE) working on AIPS++: Dick Crutcher (5%), John Pixton (50%), Harold Ravlin (50%), Doug Roberts (75%), and Dan Briggs (100%). In addition, NCSA is funding the salary of Wes Young at NRAO-Socorro and will shortly fund the salary of a second NRAO person. Accomplishments during the past quarter are described below under each person.
He served as local AIPS++ project manager and as BIMA director in charge of our participation in AIPS++. The job of local AIPS++ project manager is largely being taken over by Doug Roberts.
He has focused on the AIPS++ display library, under the direct supervision of Neil Killeen. He will be leaving NCSA and the AIPS++ project this summer.
He administers our AIPS++ workstations and maintains and improves aipsview. His aipsview work included the following:
He also attended the NCSA Alliance '98 conference.
Doug Roberts has been working on using the optimized FFT routines in the SGI/Cray Scientific Library (SCSL), available on the SGI Origin 2000. The SCSL may become available for other platforms in the next year or so. The results of the FFT implementations is that on a single processor the the SCSL is faster than the FFTPACK version by an average of 50 in a multiple processor version as well, which should show significantly more speedup on large problem sizes. The SCSL has built-in multidimensional FFT's 2D, 3D, and N-dimension. Transforming a 2-dimensional matrix using the built in 2D transform does not appear to be significantly faster (within 10-20 into 1D FFT's. Doug is going to change the FFTServer code to use the SCSL on the SGI's.
Doug Roberts has installed AIPS++ on the 4-processor SGI Onyx at the astronomy department and on his SGI Octane. The SGI systems have built using the native SGI MIPSpro compilers. On the Origin system, a red/black system has been set up to allow parallel development on a stable system while code fixes are tested in a separate version.
As noted above Dan Briggs was killed in a sky-diving accident on July 5, 1998. We decsribe here his work over the last quarter. Dan Briggs continued to investigate parallel gridding. Dan Briggs started design work on extending the aips++ task 'sky' to include full wide field imaging capability. Dan rewrote chapters 6 and 8 of ``Synthesis Imaging in Radio Astronomy'', one of the fundamental references for new AIPS++ users. Dan traveled to Socorro June 16 through June 23 to present imaging lecture at NRAO summer school and to work with Brian Glendenning and Tim Cornwell on wide field imaging design. Dan installed Classic AIPS on NCSA SGI workstations for comparison use against AIPS++. Dan made minor bug fixes involved with the port to the Origin 2000. He took a more active role in assisting with AIPS++ system administration, in support of getting a reliable port and installation on the Origin 2000. Worked on Local Organizing Issues for the upcoming ADASS meeting.
Local project members: Ger van Diepen (GVD), Michael Haller (MXH), Jan Noordam (JEN, local manager).
The elevation of Ger van Diepen to Interim Technical Leader from September will make things more difficult locally at NFRA. However, its is probably the most effective thing to do in the short term, and various Good Things may come from it.
Athol Kemball has made a very fruitful visit to Dwingeloo in June. All the more so since it coincided with the yearly visit of Wim Brouw.
Bob Campbell (postdoc at NFRA) has made a large contribution to the definition of ionospheric calibration in AIPS++. He will remain involved in the near future.
This quarter, the NFRA contribution in FTE's was as follows:
| Global AIPS++ | Local AIPS++ | Other | |
| Ger van Diepen | 40% | 50% | 10% |
| Michael Haller | 50% | 40% | 10% |
| Jan Noordam | 30% | 40% | 30% |
Commissioning of the new WSRT backend (DZB) and on-line software (TMS) is going more slowly than expected. This has made it easier for JEN to keep pace with developments in providing local AIPS++ applications.
Actual data reduction still takes place in other packages (NEWSTAR, AIPS, DIFMAP). The various data conversion programs are sufficient for the moment, but would have to be upgraded for more advanced observing modes. Therefor, processing in AIPS++ itself remains a high priority.
In the context of writing local applications for the WSRT upgrade, JEN has created a number of Glish scripts that might be useful for other applications. They include a 'standard' application interface consisting of a menubar, a textwidget, a statusline and a dismiss button. This has proved a rather useful concept in practice, espacially because of the various parameter-input services of the menubar object. In addition, there is a general tool for inspecting Glish variables of any shape or type (including tables), and an on-line help summary of the Glish language.
MXH's work on DatalineTool (user of pgplotter for visualiser) will make use of the application interface sketched above. It will have many more possibilities for user-interaction with the plot than its gplotid-based predecessor xyftool.
The progress of the HOLOG (holography) program by GVD was hampered by the necessity to add support of system temperatures in the conversion programs.
The conversion program ms2scn has been extended by GVD to take the system temperatures into account.
Class UVFITSWriter has been changed by GVD to optionally write the system temperatures and gain curves to a UVFITS file. It makes it possible to import these aspects of a WSRT measurementset in classic AIPS.
Harro Verkouter (JIVE) has taken on responisbility of a more permanent version of the UVFITS writer that will alos allow uv-data calibration in AIPS (rather than just imaging).
GVD has worked with Killeen to fully support regions/masks in the lattices and images. On 15-16 April Neil paid a very fruitful visit to Dwingeloo. The amount of email messages exchanged must be enormous. It's a good thing Neil works late, so our working days overlap a great deal. A class MaskedImage has been introduced to make the inheritance tree for SubImage and ImageExpr better. The LCRegion classes have been changed a bit (e.g. for better round-off) LatticeApply and associated classes have been changed to make use of regions/masks possible by operating on MaskedLattice instead of Lattice. Also a few other improvements have been made to these classes. Neil uses them in his Moments and Statistics classes. The LatticeExpr classes have been changed similarly. They operate as best as possible by only taking a mask into account when present.
Apart from a few small bug fixes and optimizations, not much has been done on the table system. The most notable bug fix is the .nfsNNN problem. When a table was accessed on an NFS-mounted disk, sometimes a .nfsNNN file appeared which prevented the table from being deleted. Finally I learnt that NFS renames an open file to .nfsNNN when it is deleted before being closed. Indeed a file was not closed in special circumstances. The table client got some new functionality to make it possible to handle table locking in glish scripts. To improve the performance of access to reference tables several XXXColumnCells were added to the data managers. They greatly reduce the calling and sometimes locking overhead. A class TableLocker has been introduced to make automatic release of locks possible, especially in the case of exceptions. For this purpose it has been derived from Cleanup.
On 4-5 June GVD has attended a PAO-course on ODMG 2.0, the new standard for object databases. It was very interesting, especially the subject of OQL (its query language). Ger is planning to introduce some OQL concepts in the table query language.
On 22 June GVD attended a day devoted to Corba. It was rather interesting. It was good to hear that Corba will support DCOM.
At least 20% of Michael Haller's time was spent on either normal maintenance or a variety of problems. One major problem here has been the aips++ documentation. In almost all cases the source of documentation problems has been differences here in documentation package versions to those in Socorro. Identifying these problems has been a rather long-winded process but the recent movement to latex2html version 98.1 inspired a focused effort to fix these problems and with the help of Henk Vosmeijer and Wes Young the we appear to have succeeded.
MXH also set up binary distribution of AIPS++ on linux machines at Westerbork and JEN's laptop.
The core NRAO AIPS++ group is now Barnes (100%), Cornwell (100%), Garwood (100%), Holdaway(100%), Kemball(100%), Marson (100%), McMullin (100%), Schiebel (100%), Uphoff (100%), Weatherall (50%), and Young (100%).
Tim:
Bob Garwood's primary responsibility is to oversee the single dish work which currently is focussed on the DISH environment as well as supporting the GBT.
Over the past 3 months (April through June) he has done the following work:
Brian Glendenning attended Alliance '98 at NCSA, and had many discussions about parallelizatin in AIPS++. It was decided to attempt a Windows NT port if a suitable candidate can be found, to enable AIPS++ to run in a parallel fashion on a Windows NT cluster. We also decided to hire a person to look into parallel I/O for AIPS++. Brian helped define an applicator/algorithm set of classes which should allow lattice based computations to parallelize very readily.
Brian made a number of small fixes to the tasking system, and in general started winding down his AIPS++ activities.
Brian gave presentations at the NRAO Visiting Committe, and at the Clark Symposium.
Time has been spent this quarter on synthesis planning and coordination. In particular Athol visited ATNF for three weeks and NFRA for 10 days in this regard. This has allowed detailed synthesis planning at all levels and coordination with the management at both consortium sites. A draft synthesis development plan incorporating these changes has been developed.
Significant effort has continued in cross-calibration development, as part of the continuing implementation of this capability for the third beta release. Discussions of calibration questions have been held throughout the project.
The MS v2.0 definition is essentially finalized and will be implemented soon after the third beta release. A convention for data selection in synthesis has been proposed and discussed.
A command-line parameter-setting shell has been developed as a prototype, and will be made available in the next release. This user interface uses the same meta-information as the existing GUI components.
In the last quarter Ralph has spent most of my time in the cleaning up, inproving, and testing the code in the ComponentModels module. This module is used to represent components of the sky brightness using a functional form. Some of the things that have been done are:
He has also spent some time this quarter examining some speed issues in the Clark clean and testing the egcs compiler as a alternative to the gnu compiler. Code cop duties typically take 10% of Ralph's time.
Joe's main responsibilities are managing the Green Bank installation of AIPS++ and supporting AIPS++-related GBT needs.
Darrell Schiebel implemented all of the Glish changes and additions described in the section on progress in the last quarter. He also worked on the GBT observer table parser client.
Jeff worked in a wide range of areas, from supporting dishplot to supporting the AIPS++ system.
He looked into accuracy questions for cos() under Linux/Intel that Wim reported. Answer: it's apparently a known problem.
He continued work on Dish/Dishplot:
He continued work on AIPS++ infrastructure/system:
He also worked on other miscellaneous tasks:
In this section, we give the names of people in the various AIPS++ groups and the nominal fraction of time allocated to AIPS++. These are as expected for the next quarter (and thus do not include the expected replacements for Briggs or Uphoff, or the new hires in Socorro).
The ATNF group is: Neil Killeen (25-75%, also local manager), Wim Brouw (75%), Mark Wieringa (30%) and now David Barnes (75%) as of April 6 1998.
The BIMA/NCSA group is: Dick Crutcher (5%), John Pixton (50%), Harold Ravlin (50%), Doug Roberts (75%).
The NFRA group is: Ger van Diepen, Michael Haller, and Jan Noordam. The allocations are:
| Global AIPS++ | Local AIPS++ | Other | |
| Ger van Diepen | 40% | 50% | 10% |
| Michael Haller | 50% | 40% | 10% |
| Jan Noordam | 30% | 40% | 30% |
Thus, in aggregate, we have 21 people contributing just over 17 FTEs to the AIPS++ Project. Of these, 9 are employed by NRAO, and contribute 9.4 FTEs. The numbers for the other partners are: ATNF 4 and 2.80, BIMA/NCSA 4 and 1.8, NFRA 3 and 1.2.
