AIPS 15APR98 In August 1997 the decision was made to incorporate Eric Greisen's Charlottesville Experimental Version of AIPS (CVX) as the next version of AIPS after 15OCT97, becoming the 15APR98 version. CVX had been kept up-to-date with the standard AIPS during 1996 and 1997, making the merging process fairly straightforward. Extensive testing of the 15APR98 version in Socorro and Charlottesville identified a a modest number of bugs in the new version, some of which were also in previous versions. Within a month or two the number of bugs turning up dropped to that normally seen for TST versions, and all incompatibilities between the versions were resolved. This successful merging was possible due to the effort made in Charlottesville to keep CVX up-to-date with developments and bug-fixes in the standard version. Some of the improvements found in the 15APR98 version are several new interactive and non-interactive data editing tools (some of which had been ported to 15OCT97 previously), improved bandpass calibration methods, improved tools for examining the quality of calibration, and enhancements to the interactive imaging/self-calibration tasks. New algorithms to allow single-dish beam-switched continuum imaging were included, as well as enhancements to the 12m spectral-line on-the-fly imaging. Solutions to the Y2K problem, remote display capability, and data files in excess of the classical 2-Gbyte limit (also available in 15OCT97) have all been implemented into the 15APR98 version. An agreement outlining the steps in a transition between AIPS and aips++ was reached in early 1998. Non-coplanar imaging Perhaps the most important enhancement scientifically in the 15APR98 version was the addition of non-coplanar imaging to the basic imaging task IMAGR. Previous imaging tasks, when making images well away from the original phase-tracking direction, corrected phases to the center of the image, but did not re-project the interferometer coordinates. The error due to this treatment goes as the product of the angle between the source and the center of the image times the (much larger) angle from the center of the image to the original phase tracking point (at P-band, this can cause the field of view over which the error is acceptable to become smaller than the synthesized beam!). The 15APR98 version of IMAGR allows a "3D" option in which the interferometer coordinates are re-projected so that all fields imaged are tangential to the celestial sphere. The error by this method is proportional to the square of the angle between the source and the center of the image. This option adds only about 1% to the CPU time of the job (when the improved imaging is insignificant) and can speed up imaging by a great deal when the improved image quality does matter. The 15APR98 version of IMAGR also allows imaged fields to overlap without loss of information and even provides an experimental sequence of cleaning which allows a source to be cleaned from more than one field without convergence problems. The number of fields allowed is 64 rather than the previous 16. Hardware performance Recent hardware performance tests yielded some promising results. These tests were triggered by two events: (1) Sun recently announced new products intended to compete directly with both modest and high-end PCs, and (2) the GNU Fortran compiler (g77) has been made to work with AIPS. Previously for PCs running Linux we converted the Fortran AIPS code to C via f2c and then compiled with gcc. The latest version of g77 does a significantly improved job of optimization during the f2c conversion and the C compilation (our theory), resulting in significant performance improvements. AIPS compiled and optimized with g77 was found to run substantially faster than AIPS compiled with f2c followed by gcc. On the standard "DDT" test, a Pentium Pro 200 (dual-processor) went from 3.6 AIPSMarks under f2c to 6.7 AIPSMarks with g77. On a well-equipped Pentium II 300MHz PC, the results went from 4.6 to 9.6 AIPSMarks. Sun's Ultra 5 (low-end) computer was found to give 4.3 AIPSMarks while an Ultra 30 gave 10.0. Both of these were seriously limited by disk I/O. When scratch files were directed to /tmp, which is actually the swap area maintained as much as possible in memory, the results jumped up to 7.6 and 15.0, respectively. A Sun Ultra-1/170 goes from 5.9 to 7.8 AIPSMarks using this trick. This trick can cause big problems if the swap space is used up. Clearly Sun's operating system is not managing disk cache as well as we had assumed. Full results of all recent AIPS benchmarking can be found via the AIPS homepage. The excellent performance of the Linux machines using g77 brings PCs into direct competition with workstations as AIPS machines. Distribution The 15OCT97 version of AIPS has been distributed to 76 sites, running 184 installations (including Solaris, Linux, DEC Alpha, HP and SGI versions). The majority of AIPS distributions are now by ftp. At present it is planned to release 15APR98 in mid April. Developments Multiple TV displays One of the restrictions that AIPS users have often encountered was the limit of one instance of the TV server (XAS) per host. Work is nearing completion to facilitate running several instances of the TV server on a single host (using both Unix and Internet socket connections). There will be some restrictions on connections from remote hosts to local TV displays, but the current functionality will be greatly enhanced. It will even be possible (for example) to have many AIPS sessions, some talking to remote hosts and some to different AIPS sessions on the local host, all with separate TV displays. Multiple TEKSRV and MSGSRV servers will also be available on the same host, i.e. more than one plot and message window can be used. This long-requested expansion to the user functionality of AIPS should be available by April in the 15APR98 release. AIPS on CD-ROM Work on a script to integrate the read-only file system on a CD-ROM to a partial or full installation of AIPS is under way; good progress has been made. Expect the first trial CD-ROMs to be written by the end of March. The "partial" installation will permit the user to install a minimum of the system on her or his hard disk, and run mostly off the CD-ROM itself. The "full" installation will put a complete copy of AIPS on disk so that the presence of the CD-ROM is not necessary in order for AIPS to run. Both of these options may be desirable for various users. DLT status NRAO purchased two identical Digital Linear Tape (DLT) model 7000 drives; these have a claimed capacity of 35 Gigabytes uncompressed, and should achieve very high throughput. Initial tests with modest and large size AIPS databases gave actual transfer rates of about 3 Megabytes per second. This is significantly faster than any other tape backup device within NRAO, excluding the VLBA correlator. One device was shipped to Socorro in early 1998 but has malfunctioned and is under (warranty) repair/replacement. The other was initially in Charlottesville, and was used to test out both a DEC Alpha and Sparc Ultra in terms of system compatibility; these tests were successful. OpenNT results OpenNT is a commercial product which provides a Unix shell operating on top of a Windows NT system. An effort to try and port AIPS to OpenNT has recently begun. The OpenNT system has been installed, but lack of many common Unix features (e.g. symbolic links) has hampered efforts to implement this port. There are many other system problems looming, and the success of this effort is uncertain. Geodetic Software With input from members of the geodesy community [Ed Fomalont(NRAO), Dave Gordon(GSFC), Dan LeBach(CfA), and Craig Walker(NRAO)], AIPS modifications are being made to accommodate processing of geodetic data from the VLBA correlator. Several hang-ups have been identified and are currently being worked on. A data-processing path through AIPS meeting the needs of the geodesy community has been identified and is nearly complete. The task FRING has been cloned into a new task KRING in which problems with memory and scratch file usage are being addressed. Further modifications to accommodate geodetic data reduction are planned. Documentation The VLBI chapter in the `AIPS Cookbook' user manual was completely revised, to include extensive documentation on the use of AIPS for Space VLBI data analysis. New understanding, developed as work with VSOP observations proceeds, is being incorporated continuously into an electronically-accessible version. General developments A Van Vleck correction has been added to FILLM. Initial tests show a substantial decrease in artifacts for very bright, extended sources. Parallelization We are currently introducing support for shared-memory multiprocessor machines into the AIPS computational libraries. FRING has been adapted to take advantage of more than one processor and shows significant savings in execution time when multiple processors are used. Our initial parallelization efforts are targeted at the SGI Origin series of computers, taking advantage of NRAO's Origin 200. Code written for SGI's machines can easily be adapted to run on machines that support the OpenMP standard. Many of the changes required to support parallel execution have also proved to be beneficial on single-processor systems so that all AIPS users can be expected to benefit from the parallelization initiative. Real-time VLA data filling The possibility to fill VLA data in real time remains a popular method for AOC-based users to inspect and reduce their data rapidly. In 1997, some new bugs were fixed, increasing the general robustness of the system. Cause for concern is the limited bandwidth between the VLA and the AOC, which remains an obstacle preventing large spectral line data sets or high time resolution continuum data sets to be filled in real time without loss of data. The August 1997 upgrade of the operating system software on the real time FILLM server at the VLA site uncovered a severe problem with filling VLA data in real-time. Small packets from the VLA real time computers would intermittently hang the visibility server on miranda and any on-going or subsequent FILLM's would not get visibility data. This problem was resolved at the beginning of November 1997. Sometimes increasing popularity of a new mode of observing uncovers problems that we were not aware of before. An example is 4 band, which may require rapid switching between continuum and 512 channel spectral-line mode. Under these circumstances, the visibility server on the Modcomps would send "bad" data over the network, causing FILLM and the visibility server process to crash. These crashes did not properly release connection slots. A data sanity check was put into the visibility server software on miranda to prevent bad data from being sent down the wire. Also the segmentation faults are now being trapped, and the connection slots are properly released when the visibility server or FILLM crash. The VLA real-time group is currently addressing the problem with the Modcomp visibility server. Space VLBI Developments in AIPS Most of the Space VLBI enhancements to AIPS had been completed and tested, as extensively as possible using simulated Space VLBI data, prior to the launch of Halca. The early fringe searches revealed a number of minor problems in other, general AIPS tasks. In particular, not all tasks properly accommodated the baseline-dependent integration intervals, a feature developed to keep the correlator's output data rate within reasonable limits while maintaining wide rate windows for ground-space baselines only. These problems were resolved quickly, and AIPS was used effectively throughout the initial fringe-search phase. Later stages of data analysis -- calibration and imaging -- appear to function normally for Space VLBI observations, at least superficially, and were used with some success to produce a number of images from early VSOP observations. Careful study of the imaging stage, including development of optimal techniques for using the AIPS imaging tasks, however, is just beginning. A high-performance computing system was procured from Silicon Graphics, Inc., for use by Space VLBI observers. This system, consisting of an Origin 200 multi-processor unit and two O2 front-end visualization workstations. The SGI system proved invaluable in evaluating both test and scientific data, from the early searches for Halca fringes, to the increasing number of scientific observations currently being scheduled. In particular, it has been possible (after some initial difficulties) to exploit its large memory capacity and parallel processing capabilities for very efficient fringe searching. It is important to keep track of the position of an orbiting antenna while reducing space VLBI data. Aside from the obvious use of the position in recalculating baseline vectors, quantities derived from the position of the antenna may be used in the calibration and editing of space VLBI data. For example, an astronomer may want to delete data taken when the satellite was close to perigee since the quality of the data is expected to be poor at this part of the orbit because the position of the satellite is most uncertain at this part of its orbit. Over the past year we have added facilities for storing the position of orbiting antennae with AIPS data. The position and velocity vectors of the satellite may be tabulated as a function of time in an orbit (OB) table, together with several derived quantities such as the orientation of the antenna, and mean orbital elements may be stored in the antenna (AN) table. Several AIPS tasks, including OBEDT, UVFIX, and VPLOT, can make use of the data stored in these tables HALCA has a relatively small dish collecting area and a high system temperature compared with ground-based antennae. Combined with the fact that sources are more highly resolved on longer baselines, this means that the signal is usually weak on ground-space baselines. In some cases it may be necessary to add the signal from several baselines (using closure relations) to find fringes to the space antenna. This technique has always been available in FRING but has also been incorporated into the baseline-oriented fringe fitting program BLING during the past year. BLING has also been made more efficient (its run-time performance has improved by more than an order of magnitude) and has been adapted to cope with the large ranges of delay and rate that must searched in some space VLBI experiments.