IT hardware and infrastructure

As might be expected, an unmanned large-scale remote-controlled facility like ALIS requires considerable amounts of Information Technology (IT), both in terms of computers and datacommunication infrastructure. As work on ALIS was initiated, much of what is taken for granted today simply did not exist, or was far too expensive. In other cases the technical evolution took unexpected turns, making some choices of technology appear awkward and obsolete today.

Computers

A variety of computer architectures and operating systems were available when work on ALIS started. At an early stage command and control functions were separated from image processing, data storage and transmission. The first specification of the computer systems for ALIS [Brändström and Steen, 1992] included a main computer and a dedicated Image Processing Computer (IPC) in the control centre. At the stations, one computer would be responsible for controlling the station, while a special dedicated unit, called the Near-sensor Interface and Processing Unit (NIPU), would handle the large amounts of data ($\ge 2$ Mbytes/s) created by the imagers.

A Hewlett Packard HP-755 PA-RISC workstation was selected as the control centre main computer. Budgetary constraints prohibited the use of similar computers at the stations. Therefore it was decided to use IBM-PC compatible machines (i486). Due to the fast development of computer hardware, the station computers have been replaced after typically 3 years of operation. The current system still uses various PCs (ranging from i486 to Pentium III). At the control centre the HP-755 computer lasted until 1999 when it was replaced by two PCs. The aim of having a dedicated image processing computer has not been realised as no real-time data is yet available due to the absence of high-speed lines to the stations. Dataanalysis is performed on various workstations.

The NIPU

The datahandling, image processing and data storage at the stations were supposed to be handled by a dedicated computer. Around 1990, a very promising device for this purpose was the Transputer, a parallel processing device communicating with other devices over four serial links. Another interesting device was the Intel I860 floating-point processor. A prototype system was built using the T222 and T800 Transputers. Apart from the imager itself it was also desirable to control the Camera Positioning System (CPS) as well as the filter-wheel from the NIPU. In this way the NIPU would control all subsystems related to the imager. The ALIS imager would produce over 2 Mbytes/s, which was too fast for the T222/T800 Transputer links (capable of 20 Mbits/s). However a new Transputer, the T9000, was expected to be released around 1992. This device would have enhanced links capable of 100 Mbits/s and increased processing power (25 MFLOPS), and would thus be well-suited for demanding image-processing applications [Pountain, 1991]. Designs were made for this device, but its release was postponed several times due to technical problems. Since the camera controller for the ALIS-Imager (Chapter 3) was also based on a Transputer, the T222, a T800 in the NIPU was chosen as an intermediate solution for handling the incoming image data. Due to the limitations of the link-speed of these Transputers, data was to be transferred to the NIPU over a 16-bit parallel interface. Six NIPUs of this design were built (Figure 2.5).

Figure 2.5: The NIPU. The board contains four T222 modules, each controlling the $\alpha$ and $\beta$ axes of the camera-positioning system as well as the filter-wheel and an additional Transputer intended (but never used) for GPS-timing. The large module is the T800-board with memory and parallel interface for image capture from the 16-bit parallel interface of the camera controller. Ample space is provided for future expansions with T9000 boards.

Meanwhile the fast development of the PCs eventually made the Transputers obsolete, and around 1997 it was decided that a second PC would take over the responsibilities of the NIPU. A prototype PCI-board with a DSP was designed and tested, but during this work technical problems with the parallel read-out from the camera controller were discovered (Section 3.3.4). Since these problems could not be resolved, the resulting decrease of the maximum imager frame-rate made both the NIPU and the second PC for receiving image data superfluous. Therefore the six already existing NIPUs ended up as rather over-designed controllers for the camera-positioning system (Section 3.6) and the filter-wheels (Section 3.5.1).

Communication systems

When design work on ALIS began, the fastest off-the-shelf modem available was at 2400 bits/s, and with a special, rather expensive, leased-line one could attain 9600 bits/s. The first ALIS paper [Steen, 1989] specified a $\ge 10$ Mbits/s communication link capable of near real-time image transfer to the control centre. A network of microwave links was considered but deemed far too expensive, as was the case with the fibre-optic lines passing near two of the stations (Merasjärvi, Silkkimuotka, see below).

Dial-up telephone lines were too slow, and faster means of communication too expensive. It was anticipated that the fast technological development in this field would make faster communication lines available at a reasonable cost. Thus it was decided that ALIS would use slow dial-up modem lines for command, status information and to transmit reduced quick-look images to the control centre. A future ``to be defined'' high-speed link to the control centre would provide the high-speed communication required for real-time transfer of raw image data. Meanwhile, local data storage, and on-site image processing of the data would be employed at the stations.

The dial-up lines were one of the major sources of trouble in ALIS during the early years. This was mainly due to bad telephone lines and old electro-mechanical telephone switching equipment. This led to extensive efforts to troubleshoot modem lines and to develop reliable communications software. Also the modem technology and quality of the telephone lines improved considerably over the years. Today the dial-up lines are capable of reliable 28800 bits/s communication, using the standard Point-to-Point Protocol (ppp) [Simpson, 1994].

The high-speed link remains to be defined. The optimal solution would be optical fibres ($> 100$ Mbits/s), but other solutions are also possible, such as ADSL ($500$ kbits/s), ISDN ($<128$ kbits/s), radio-links ($>1$ Mbits/s), etc.

Stations (2) Merasjärvi and (3) Silkkimuotka are located in the proximity of nodes for high bandwidth fibre-optic communication lines. (5) Abisko is located close to the Abisko Scientific Station (ANS) which recently acquired high-speed fibre optic Internet connection.

Presently only the Kiruna station has a 2 Mbits/s Ethernet connection to Internet realised by a microwave-link to IRF, the rest of the stations are connected by means of 28 kbits/s dial-up modem lines. The rapidly increasing demand for high-speed Internet subscriptions among the general public might speed up the process of getting faster communication lines for all ALIS stations at a reasonable cost.

Station data storage

Various solutions for the local data storage at the stations have been considered over the years. Initially it was intended to store the image data onto Digital Data Storage (DDS) tapes which around 1992 had a storage capacity of up to 1 Gbyte. However, this solution proved slow and unreliable, mainly due to the hostile environment at the stations during tape-changes (moist, rapid temperature changes, etc.). Other solutions were also studied, but most of these were too complicated, too expensive or both. If faster communications would have been available, data could be stored on hard-disks, and downloaded to the control-centre in near real-time, or during non-measuring time. As the DDS drives tested at the first stations were not as reliable as expected, large (i.e. 2-9 Gbytes around 1992) external SCSI hard-disks were used instead. When a disk became full, it was exchanged manually, either by neighbours to the stations, or by staff from IRF. This solution proved simple and reliable. The only disadvantage was the usually rather long time (typically months) before raw-data from all stations became available for archiving and analysis.

Data archiving and availability

Reduced quick-look images (about 16 kBytes) are transmitted to the control centre and distributed to the operations centre (and web-site) in near real-time during measurement. However, these images are only intended for monitoring, and are of far too poor resolution for scientific analysis.

As the raw-data disks reach the control centre, recordable CDs (CD-R) of ALIS data are produced, and archived. All ALIS data produced so far are also made freely available on the world-wide web (see http://www.alis.irf.se for details) The main archive web-site is maintained by Peter Rydesäter who also provides a SQL database and search tools (see also Section 6.2).

The image data is stored in the Flexible Image Transfer System (FITS), [NASA, 1999]. This format is in wide use by the astronomical community, and found to be particularly suited to store scientific image data, as all supplementary information regarding an image (exposure time, filter, CCD temperatures, subsequent processing etc.) can be stored in the image header in a flexible way. FITS is recognised by many image processing packages, and free conversion programs to most other image formats exist on the Internet for most operating systems.

The size of the image-files is 16 bits/pixel (2 bytes) where the number of pixels is dependent on the configured spatial resolution (see Table 2.2 and Section 3.2). The total size of a set of images is also dependent on the number of stations involved and the temporal resolution selected.

Operating Systems

It was an early requirement to have a true multi-tasking operating system such as Unix for ALIS. HP-UX, which was delivered with the workstation selected for the control centre fulfilled this demand. The decision to use the IBM-PC architecture at the stations limited the choice of operating systems to SCO-Unix and MS-DOS. SCO-Unix was quite expensive compared to its reliability, so the reluctantly chosen remaining option was to use MS-DOS at the stations. This lead to limitations in the flexibility of the system.

Some years later highly reliable and free operating systems such as Free-BSD and GNU/Linux emerged. It was immediately realised that a change to one of these operating systems at the stations would be necessary to meet the required data-handling specifications. In 1997 all stations had changed operating systems to Debian GNU/Linux, and in the fall of 1999 the HP-UX operating system in the control centre was also changed to GNU/Linux as the old HP workstation was replaced.