+ All Categories
Home > Documents > Using Open-Source Software and Low- Cost Computers for ... · Cost Computers for Earth-Observation...

Using Open-Source Software and Low- Cost Computers for ... · Cost Computers for Earth-Observation...

Date post: 18-May-2018
Category:
Upload: trinhkhanh
View: 213 times
Download: 0 times
Share this document with a friend
5
64 r bulletin 103 — august 2000 Using Open-Source Software and Low- Cost Computers for Earth-Observation Data Processing M. Önnestam ESA Directorate of Applications Programmes, ESRIN, Frascati, Italy W. Balzer Deutsches Zentrum für Luft- und Raumfahrt (DLR), Oberpfaffenhofen, Germany M. Curtis National Remote Sensing Centre Ltd. (NRSC), Farnborough, United Kingdom Processing Earth-observation data The Agency’s Earth-observation projects involve space missions in which instruments carried by polar-orbiting satellites provide data to numerous users all over the World. This data arrives in a raw format requiring processing on the ground to convert it to data products that can be understood and exploited by users. This processing has until recently required very specific hardware and software, implying expensive electronic and computer equipment. The evolution of technology over the last few years, however, has now made it possible to use standard, high-performance computers even for these demanding tasks. System re-hosting The current evolution of technology in general and computers and electronics in particular, means that systems that were fully state of the art in their design when built rapidly become obsolete as more modern solutions become available. This creates problems in maintaining equipment, as spare parts for and knowledge of the systems become scarcer after a number of years. It also becomes increasingly difficult to add or change functions of the original systems, as their performance is limited and difficult to increase. It can also be difficult to find new peripheral equipment compatible with the older technology. The basic algorithms used for data processing meanwhile often need to be maintained in the same state throughout the lifetime of an Earth- observation mission, which can last for many years. The in-orbit instrumentation always remains the same, although it may degrade slightly. This means that the original software and algorithms need to be kept for several years, whilst the computer hardware needs regular replacement. The process of changing the computer equipment while maintaining the software and algorithms is called ‘re-hosting’. It means that hardware is replaced by new equipment while the software remains, wherever possible, identical. The main problem in this process is that new hardware often requires a new intermediate layer, the Operating System, where basic input and output and most importantly mathematical functions are implemented in different ways. This leads to situations where the original software needs to The introduction of open-source software compliant with industry standards has allowed the use of low-cost standard computers for a new set of Earth-observation applications. The Linux operating system is an example of this kind, which makes it possible to use ordinary PCs for demanding data-processing needs. ESA has benefited from this opportunity by replacing old and difficult to maintain systems with new ones providing both a significant increase in performance and a reduction in investment as well as maintenance costs. The Agency utilises a number of facilities in Europe for its Earth-observation data- processing tasks, where equipment, systems and resources have been assembled to form entities that provide the required services. The ERS mission is such an example, supported by so-called ‘Processing and Archiving Facilities’ (PAFs) where the data are archived and made available for processing. The PAFs have data- processing systems that read data from the archives and produce data products for the users.
Transcript

64

r bulletin 103 — august 2000

Using Open-Source Software and Low-Cost Computers for Earth-ObservationData Processing

M. ÖnnestamESA Directorate of Applications Programmes, ESRIN, Frascati, Italy

W. Balzer Deutsches Zentrum für Luft- und Raumfahrt (DLR), Oberpfaffenhofen, Germany

M. CurtisNational Remote Sensing Centre Ltd. (NRSC), Farnborough, United Kingdom

Processing Earth-observation dataThe Agency’s Earth-observation projectsinvolve space missions in which instrumentscarried by polar-orbiting satellites provide datato numerous users all over the World. This dataarrives in a raw format requiring processing onthe ground to convert it to data products thatcan be understood and exploited by users.This processing has until recently required veryspecific hardware and software, implyingexpensive electronic and computer equipment.The evolution of technology over the last fewyears, however, has now made it possible touse standard, high-performance computerseven for these demanding tasks.

System re-hostingThe current evolution of technology in generaland computers and electronics in particular,means that systems that were fully state of theart in their design when built rapidly becomeobsolete as more modern solutions becomeavailable. This creates problems in maintainingequipment, as spare parts for and knowledgeof the systems become scarcer after a numberof years. It also becomes increasingly difficult toadd or change functions of the originalsystems, as their performance is limited anddifficult to increase. It can also be difficult to findnew peripheral equipment compatible with theolder technology.

The basic algorithms used for data processingmeanwhile often need to be maintained in thesame state throughout the lifetime of an Earth-observation mission, which can last for manyyears. The in-orbit instrumentation alwaysremains the same, although it may degradeslightly. This means that the original softwareand algorithms need to be kept for severalyears, whilst the computer hardware needsregular replacement.

The process of changing the computerequipment while maintaining the software andalgorithms is called ‘re-hosting’. It means thathardware is replaced by new equipment whilethe software remains, wherever possible,identical. The main problem in this process isthat new hardware often requires a newintermediate layer, the Operating System,where basic input and output and mostimportantly mathematical functions areimplemented in different ways. This leads tosituations where the original software needs to

The introduction of open-source software compliant with industrystandards has allowed the use of low-cost standard computers for anew set of Earth-observation applications. The Linux operatingsystem is an example of this kind, which makes it possible to useordinary PCs for demanding data-processing needs. ESA hasbenefited from this opportunity by replacing old and difficult tomaintain systems with new ones providing both a significant increasein performance and a reduction in investment as well as maintenancecosts.

The Agency utilises a number of facilities inEurope for its Earth-observation data-processing tasks, where equipment, systemsand resources have been assembled to formentities that provide the required services. TheERS mission is such an example, supported byso-called ‘Processing and Archiving Facilities’(PAFs) where the data are archived and madeavailable for processing. The PAFs have data-processing systems that read data from thearchives and produce data products for theusers.

ONNESTAM 22/8/00 11:04 am Page 2

65

low-cost eo data processing

be updated although the intention is to keep itunchanged. It also means that thoroughvalidation and testing is required to prove thatthe re-host generates output indistinguishablefrom that of the original system.

Open-source softwareThe recent evolution in networking using theInternet has enabled software developersaround the World to co-operate in the so-called ‘open-source’ movement. This has beenbased on the GNU General Public license forsoftware whereby anyone has the right tomodify the code and re-distribute it as they seefit. The Linux operating system is one exampleof this international co-operation. It istechnically a Unix kernel managing thecomputer hardware and networks that wasfirst developed for the PC architecture. It isnow also available for many other platforms.

It is possible for anyone, through the licence, to acquire the source code of Linux bydownloading it from an Internet server. Inpractice, however, is it more convenient to buya CD-ROM on which the source code isdistributed together with support programsand applications.

The fact that the source code is open andfreely available makes modification andtailoring for a specific application possible. Thecode can be changed and re-built for thespecific environment by the teams building theapplication. This means also that theapplication when ready is more amenable tofuture re-hosting, if required.

The GOME Data Processor re-hostingThe GOME Data Processor (GDP) at theGerman Processing and Archiving Facility (D-PAF) in Oberpfaffenhofen generates dataproducts for the monitoring of the Earth’satmosphere. This processor has been inoperational use since the launch of ESA‘sERS-2 Earth-observation mission in 1995. Theoutput products derived from the GlobalOzone Monitoring Experiment (GOME)instrument provide global information aboutthe atmosphere’s content of trace gases likeozone.

The GOME algorithms and processingsoftware used by the GDP have evolved overthe years since the start of operations, due bothto increased knowledge of the instrument’scapabilities and to innovations in processingtechniques. This led to data acquired atdifferent points in time being processed withdifferent versions of the data processor, leadingin turn to small but significant differences in theproduct output. However, the scientific users of

the products looking for global trends andchanges ideally require a coherent data setprocessed with the same software versioncovering the full mission. The requirement forre-processing capabilities was thereforeacknowledged by the Agency during 1998.

The GDP at that time was coping successfullywith the processing of new data, but could notsimultaneously support extensive re-processing. ESA therefore asked the institutionhosting the processing facility, namely theDeutsches Zentrum für Luft- und Raumfahrt(DLR), to examine the possibility of upgradingthe processing capacity. It was soon apparentthat there were two possibilities, to extend thesystem with the same brand of hardware or toreplace the system with low-cost PCs runningLinux. The second possibility offered a price/performance ratio an order of magnitude betterthan the more traditional first alternative.

The fact that the original processing systemused hardware running a Unix operatingsystem helped at this stage when the effort tore-host was analysed. Linux, being a flavour ofUnix aiming for standard compliance,supported the application well without requiringmajor coding changes. The Agency thereforefinally selected the PC-based solution wherebytwenty identical units would be assembled intoa ‘PC-farm’ to execute the most demandingparts of the data processing (Fig. 1).

The original software already supported anarchitecture in which the processing could bespread over a network of computers. The mainadvantage of this was that one unit could failwithout disturbing the others and that theoverall performance would only be slightlydegraded.

The scope of the project at this stage was:– Re-hosting of the most demanding part of

the application, the data-processing engine,starting from calibrated instrument data andending with user products containing geo-physical information.

– Maintenance of all existing externalinterfaces by keeping the system front-endas it was without changing the hardware.

– Implementation of memory handling tohandle the differences in processorarchitecture correctly.

The GDP re-hosting project now consisted oftwo major activities related to the applicationsoftware: the original software had to be re-builtfor the new Linux environment, and thedifferences in processor architecture had to behandled correctly. The software re-builddemonstrated the fact that Linux and software

ONNESTAM 22/8/00 11:04 am Page 3

66

r bulletin 103 — august 2000

showed very small relative differences, all withinthe level of mathematical precision, but veryfew cases where the relative difference washigh but the absolute difference was low,because the computed values were close tozero.

The D-PAF now uses the re-hosted GDP for itsGOME data processing. The first re-processingcampaign took place during autumn 1999. Theprocessor generated user products for all ERS-2 GOME data over a number of months,achieving a speed whereby one year of datagot processed in one month, despite severalhardware problems. This can be comparedwith the previous performance of processingtwo months of data in one month with nooperational failures.

The UK-PAF re-hostingThe United Kingdom Processing and ArchivingFacility (UK-PAF) has provided data-processingand product-generation services for the usercommunity since the first ESA Earth-observation mission, ERS-1, was launched in1991. The products have been derived fromthe source data of the main ERS instruments,i.e. the Radar Altimeter (ALT), the SyntheticAperture Radar (SAR) and the Along-TrackScanning Radiometer (ATSR). The architectureknown as the Earth Observation Data Centrewas first assembled to support the ERS-1mission. It was subsequently extended during1994 and 1995 to support ERS-2.

One of the main UK-PAF data products is theRadar Altimeter Waveform Product (ALT.WAP),which contains the processed RA telemetry aswell as important annotations such ascorrection factors for atmospheric conditions.This is a global data product offering a uniquedata set covering oceans, ice- and land-masses, which no other Radar Altimeter dataset is capable of delivering. The ALT.WAPprocessing chain is located only at the UK-PAFand its hardware has not been updated sincethe ERS-2 launch.

In 1998, the Agency began replacing obsoletehardware in its Earth-observation data-rocessing facilities, while also ensuring year-2000 system compliance. The UK-PAF host,the National Remote Sensing Centre Ltd.(NRSC), carried out a study of the implicationsfor their facility. It was found that thearchitecture from 1995 utilised computerhardware that was obsolete such that it mademaintenance very expensive and thereplacement of peripheral equipment likemagnetic tape devices almost impossible.Moreover, the operating system was not year-2000 compliant. It was also recognised that

designed for Linux follows accepted standardsmore strictly than other commercially availableproducts, where machine-specific extensionsare often introduced. The tool translating thecode to machine instructions, the compiler, inthis case showed this behaviour. The originalcode used many of these specific extensions,which the Linux compiler did not support. Acomplete inspection of the code was thereforeneeded to remove all such extensions. Thisalso had the beneficial effect that severalpreviously undetected errors were spotted andalso corrected.

The second major task was the implementationof the memory handling, the so-called ‘byteswapping’. The new PC architecture differed inits memory organisation from the existinghardware. The project therefore implemented abyte-swap routine at an early stage thatcorrected for this.

After these two activities had been completed,the project proceeded smoothly and theapplication software could be compiled andexecuted on the new PC hardware just 60 daysafter the kick-off. Comparisons betweenoutputs from the new and the existing chains

Figure 1. The Linux farm forGOME data processing at

the German Processing andArchiving Facility (D-PAF)

ONNESTAM 22/8/00 11:05 am Page 4

67

low-cost eo data processing

Figure 3. The schedulersubsystem of the UK-PAF.The black unit on the left isthe RAID unit, providing fulldata redundancy

critical subsystems did not have anyredundancy, which would cause long periodsof unavailability in case of failures.

All of these facts led to an NRSC studywhereby the possibility of an upgrade wasconsidered. The alternatives were soonrecognised, as in the case of the D-PAF, aseither replacing the hardware with a modernversion from the same manufacturer, but stillusing the same basic software, or replacing thehardware with PCs running Linux, which wouldalso imply system re-hosting. The fact that theoperating system in use at the UK-PAF was anolder version complicated the alternative of asimple hardware replacement, as the existingoperating system would not execute on anynew platform available. This meant that bothalternatives would effectively lead to a re-hosting, and the study showed that the effortrequired would be of the same magnitude inboth cases.

The success of the D-PAF re-hosting, togetherwith the significantly better price/performanceratio of the PC-based solution combined withmuch lower maintenance costs, led to theselection of this alternative and the re-hostingproject was started in late spring 1999. Itincluded all external interfaces, the control ofthe SAR and ATSR processing chains, and thefull ALT.WAP processing. The SAR and ATSRprocessing chains were not included as theyare delivered by other entities for the UK-PAF’suse.

The project elected to maintain the existingarchitecture, whereby discrete functions arestatically distributed over a network of

computers (Fig. 2). The lower hardware costallowed generous redundancy and eachfunction could thus reside on at least twocomputer hosts. This, combined with the useof RAID (Rapid Array of Independent Discs)technology allowing secure data storage andredundant local-area networks, removed allpossibility of outages due to a single failure (Fig. 3).

The re-hosting project not only aimed atreplacing obsolete hardware, but also had thegoal of securing UK-PAF operations in the year2000. This meant that the schedule had to bevery ambitious, to allow startup before the endof 1999. The project was therefore divided intothree phases, with the most critical functions tobe addressed first, thus securing at least limiteduse. The first of the phases secured the

Figure 2. The systemarchitecture of the UKProcessing and ArchivingFacility (UK-PAF)

ONNESTAM 22/8/00 11:05 am Page 5

68

r bulletin 103 — august 2000

Figure 4. One of the UK-PAFdata-processing sub-

systems. The unit on the leftis a robotic cassette-

handling facility enablingcontinuous data processing

difficult to understand. The values in the outputproducts differed from those of the originalchain in a way that could not be justified. Finally,a single instruction, namely a numericaldivision, appeared to be critical where therounding systematically differed compared tothe original system. This error was thenpropagated throughout the chain to the outputresults. The difference could thus be explainedand accepted and the processor becameready to use.

The UK-PAF is now fully operational for allproduction chains ten months after projectkick-off. The new systems perform accordingto expectations and well above the previoussystem levels (Fig. 4).

Re-hosting achievements and benefitsThe two Linux re-hosting projects described inthis article have both shown the excellentperformance of this low-cost alternative. It isevident in particular that the possibilities ofmultiplying the hardware in a way thattraditional alternatives do not permit raises theperformance significantly. This means in theend that much higher throughputs can beachieved for a similar or even a lower cost.

The re-hosting to Linux on PCs requiresspecific attention to two issues: the differencesin memory organisation between differenthardware types, and the differences inmathematical precision. Re-hosting projectsshould therefore identify and establishstrategies to handle these issues in the earlystages, in order to avoid surprises when thenew outputs are compared with those of theexisting systems.

The two implementations have alsodemonstrated the reliability of the Linuxoperating system. There have been very fewreported outages and all have been related tohardware or other components outside thecontrol of the operating system. Projectsimplementing this technology should thereforeselect hardware with a good reputation forreliability in order to benefit fully from the goodperformance of the Linux operating systemitself.

All things considered, Linux is proving a verygood candidate for use as the operatingsystem of choice in Earth-observation data-processing facilities. r

external interfaces of the UK-PAF to enable atleast manually controlled data processing. Thesecond integrated the functions of controllingthe SAR and ATSR processing chains, thusallowing automated processing for thesesensors. The last provided the fully automatedALT.WAP product processing.

The project experienced some initial delay dueto problems with one of the Custom Off-the-Shelf Software (COTS) products being used.This item provided the management of the UK-PAF databases and had just been released asa native version for Linux. The project initiallysuccessfully used a preview version that wasnot officially supported, but encounteredproblems when trying to use a later supportedversion. These problems proved so serious thatthe preview version was actually recovered andused in the final operational system.

The first two phases, allowing SAR and ATSRproduct generation and supporting all externalinterfaces, were finished just in time forChristmas 1999 and the UK-PAF successfullycompleted the year 2000 roll-over. The finalphase did, however, encounter problems thatdelayed the re-starting of ALT.WAP productgeneration.

The ALT.WAP re-hosting differed from the restof the project in that it touched upon the dataprocessing itself, and not just the processingcontrol. This meant that the issues of memoryorganisation and mathematical precision alsoencountered in the D-PAF case became valid.The first of these caused some problems anddelay, but were resolved quickly in a similar wayto the D-PAF when a general byte-swappingstrategy was implemented. The latter, on theother hand, created problems that were more

ONNESTAM 22/8/00 11:05 am Page 6


Recommended