Merken

Flow computation on massive grid terrains

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Beta
Erkannte Entitäten
Sprachtranskript
the bottleneck at annotation that user FIL
computational on metadata during by our Tolman not harder and they although it is prevented by developed on and but
and and around the end of the the the work that I done in my life the elementary we're both and you can hear the cannon army that arise from undertaking on L and a quarter of are monitoring flow and made got baby parameters that people have mentioned already are in that talk namely flow direction and duration but the flow real direction of that now you know that there are monitored the way water would flow in word onto that and now imagine you have 1 unit of water on each value grade and that water the water flow according to the flow direction of the problem and to compute for each cell the amount of water that flow through that that and that's called the flocculation of that so the product of environment of the ML conjured flow direction and flocculation for every ton indirect and that they are but they still behaving parameters are even tho they're are very simple they give a lot of information about the topography of the terrain off the direction of print and digital or it can be a little watershed flocculation can be a little are derived from the network and they have many many other applications which I won't even mention because I'm sure you are not a much better than me all our ground and be indicated on our very large domains which have been increasingly become available over the past few years at the lower and lower resolution I'm not operating in of the good Florida of during data on nation that I know of firepower PM it was launched last year are deleted generated January it collected on IBM they type 30 meters vision for the entire interval all types of and yet get in another part of the power thank you feel that you didn't get the proper for you at the time and the 3rd position I'm not sure if that's true anymore off and very and the that library how would you go down to 1 meter resolution a all what they call a bigger and better and better than you can have profound impact on the fact that that we have to work on and on compute and so on imagine that they pointed the Appalachian Mountain data that which is on a mountain range in United States at time in template and 100 meter regulation of without any compression thereafter both of about 500 mega right now and there came into revision it 10 times bigger than the fighting of add up and mineralization in video and at 1 meter the unit fight so that a lot of data next question is can be Matthew they have the property that software that currently in there and all according to experiments that we did all of last year well you can propagate in the patient enough we on the Net decremented experimented graph by the energy imparted by the software which compute flow direction and communication but they're older we have in the part of the modification of grandpa for instance the most commonly used on the functioning of watershed we elected 117 day planner that approximately 7 thousand by 4 thousand read are and it didn't finish I think it finished about 55 per cent in the 17th tardy and if the fate of program developed by our data problem and his group at University of Utah on it becomes available on our than expansion to argue fall all it could handle in a couple of days the the data that above that governed by by 4 thousand however we ran it on a 5 times bigger data that of about 2 thousand by 10 thousand in cell and so we added 120 days and then do because the podium finish all the time that the computerization what pertinent medical pressure and finally RT informed about flow direction and population function on it does among them so that the link there the motivation it could handle that data that in a couple of of the problem I is that it cannot handle anything that bigger than 2 gigabytes it gets out with the help of narrative and thought experiments were done on the same machine at knocking out planning it off it the pendulum with 500 megahertz 1 gigabyte of ram so and in our previous work we developed a set all programs and intercultural flow for computing the direction and plot simulation and made on they are being done theoretically optimal algorithm and they are quite a patient up to a thousand times that there are very large read them all I did top there on it scaleable that logic there and that they have read about it quite in that they that they need a resolution in both about 1 and a bit more than 1 billion billion adamant that demand it completed but that we on that and it also more effectively than the other thing that that that it allowed the user of option for choosing the type of routing and I'll get back to to that in a couple of slides but it is not the work what we're reporting and they are conference is operating the optimum to ground all function called our daughter and so we have all partially on on augmenting the uploaded additional features in order to stop you on the functionality of the other numeric graph FIL routing function on are those features like output in that sort of depression topographic convergent in that weather that query whether that data and then we have quite a readout they've done on comparing the medulla graphical routing function both in terms of output and performance but they it I'm going to talk about offered by a talk about the and what what it can be again why didn't happen automatically for program which works fine for my employer and I'll talk about that in very briefly about the theoretical approach to improve accountability which is unknown on interactive computer science and the area efficient algorithm it is not on underlying theory for the optimal then I but then I also talked vaguely about the outline of the outflow and time permitting about them all I'll go over that the graph will function and should from operationally compared very that OK on program are in general latent beginning that they it in memory however when working with metadata they had not fit in memory of the program is run because the operating theater innovative it to the user the what they that and then they get it in and out main memory and demanded by the program execution and they are all individual invisible to the user that 2 point on that 1 day moved in block between main memory and In other words if any of the program just 1 1 attracted 1 item of data it gets the entire block from that is that contain that item and 2nd thing that it did at least 3 orders of magnitude lower than acted in main memory because of the when protecting metadata because the bottleneck is not meant to be a computational but it acted thing that is the so called the ideal and the reason for that this bottleneck is that applications are normally not all aware of of their daily pattern that factor and the weight and activity that they died not really
important for much of the development of that you have looked at and then the final grade is big enough to hold for element oracle block that they have been very taking on this and I'm going to algorithms that have In the friend weighted direct the 1st algorithm activity in the order but the numbers are not the values of the area by the ordered item but at the gathering acted in the current item block it brought into main memory any other animal that had enacted in the 2nd item at 14 a different block of a block is brought into main memory the 3rd item fall in a different block the block again might be brought into main memory however the main memory for now the 1 that had to be evicted typically the video so the other without this market brought in and so on it continued by the time it reaches the peak mn they blocked the and the 5th line but in the 1st block then that have been already evicted from and I don't have to be reloaded I'm going to follow the example of the and we see that every single item that is active collided other block to be loaded from disk into main memory and now you a slightly different way in which effectively inquiry but in a different order instead not the 1st item quite in the 1st block to be noted that that an item but in the same block the block the already in memory ejected per frame the 3rd item in here then block is loaded into main memory the 4th item it in the same block at the end of the active and on the and that they are in the same block the unity that with this algorithm every block is
loaded exactly what about in implemented
all the relevant quantity in effect and 1 of the and over big in their mind number be small it will on reality can a a a larger which means that the current parliament in the factory of a thousand times it the factor of thousand times more being blocked into main memory but I can but they neglect nation for non probability the fact of clothing this block can be quite in here and they get out more detailed look into that and think that experiment that I talked about his slide that they find you with our blood watershed we're not modeled watershed on data that the ranging from a Taliban that probably has an or 1 . 8 million element in bold all thought and they that well we couldn't find here today that the 7 thousand by part about it but in the data that in this small minute no problem and the dog that the get larger 5 days and then send it there here estimated
that In the effect of all of the data item factor and the conclusion is that however good the operating system it cannot change the data type a pattern of the problem tho the the probably reading not to take care of the data that back and then it will be not scaleable but the theoretical approach for Taliban all that is so redesigned completely the algorithm in order to optimize the number of blocks that are loaded all what data that are transferred between the and main memory and they did all known as the the area I efficient algorithms the basic idea i to load each block ideally just 1 time and all the computation on on element in that block while the block is in them so can never underlying the optimal I that think that the complexity of an idea of the nitrogen either need not only that if you can't complicated but are made a number of love that I can coming back to the problem that we want to solve it in that outline all right back and getting that no direction they know that computed on and made it by looking at the by the window around the current on and there 2 options in the on a kind of direction to the detailed balance neighbor or find the direction to all the the balance of naval admitted known as being of no direction model a thinking the magical direction minority infinite on now tell you actually can be considered in 4 point which have down neighbor just look at the the by the window for the problem at the advancement of that area and on that day that you cannot compute the projection of a Belgian by looking at the neighbors because the neighbors of the thing I don't have to rout logo globally on the entire test area and we distinguish between 2 types of that area that the the matter that the flat area that if you pour water on them what good you about yeah so the 2 and then the the bottom of depression for the thing if you pour water on the type of that area but it doesn't go down now there no I'm getting direction on a direction on that area from all around the world all they by finding that each belt float wired that that close there is a point of that of that matter and so what a globally rounded toward the viewpoint of the that and things aren't option is that the only area of the only point for each direction have not been and find it odd that thing 1 thing there there's 2 options in there the ordering find the thing and I let it go out of nuclear finding loaded direction toward the center of the city and that they demand that they might act on the other option which is more theoretically interesting and all but in the referred in the literature is that of routing water outside the thing you think up here in the direction and that is normally done by stimulating flooding the sliding gradually here that statement pockets and basically the end of flooding of every point in that area and had the down the path the outside there's no thing and we can find don't of flow direction on the flooded terrain and map them back on television and so they can the idea there's nothing you in all the steps that the upload them all let me talk about documentation for a 2nd all FIL clocking up to their entire computed on population invaded by the prospect of of counting how much what about 3 point at the beginning that initially 1 in the water end of the gotten thing is that all the amplitude that there are no it's not modeling all they have been already known on the bottom the quantitative aspect we can go that I patiently baby modeling them as graph problems for geometry problem so far in its current form are but there are floating around the input
layer and innovation green and output the flow direction made flocculation topographic convergent index and the plot and depression grid they and had the option of should then between David independently what direction and for collaboration there is the option to reach to about direction routing want the full but they that pertinent threshold that the user can define and so they could call it looks like note here that the a memory main memory value that they have the have given the input data is the amount of main memory that the that all kinds by corrupt well so ideally the rich and I might depend on a number of being used they the basically had to on under and he made the amount of main memory that's available on the machine and then give that value of the parameter these are the on FIL functions that are available in ground and that the most commonly used art what had on was implemented that all you have a it takes an input and elevation output flow direction flocculation and a lot of other parameters including watersheds in that man will blank and fitness on it that humans are the brain and about whether all that I want to talk about them and the part of the deal which the hidden binary then it take the input of the network and Bainbridge network I'm not sure what that is but it had to be digitized by hand I couldn't try remember this 1 on it but it is what it has and there are about bottle of modern top idea on it in a way that top model output flow direction it elevation PPI wanted him on the problem is that it takes as input the top model parameter by which I'm not really sure what that I I don't have any better than that there and finally the following but on the floor which is applied different model and the more complex models which i are don't know much about like water at the a or hydrotest to D but I think there are more complicated there are in the future of a all the only by end that that I have honor no operated are the watershed an article well at the time that we have on the thing with that you've found them you've already been ranging from about number power and 4 thousand by Feldman and they're not 11 thousand by 11 thousand then In other words that from about 1 . 8 million adamant about 120 community and on and on and experiments were done just to make go have a much better machine now it depends on what they would have cross the 3rd 1 beloved h and that 1 day of item parameters OK but there are no taken about from 2 minutes 5 minutes and then minute up to that point 5 our all letters that they 998 women invading our land then on the data mining let it run for the data in the country Bell 5 parallel from picture it is the pandemonium the population completed by proper all you can optimize so I don't think and that is going on in my opinion is not my people direction of the population considered in light of direction in the directions matter two-dimensional picture and the main areas that at the end the of radical right at end the environment 5 gallon that's that's something found that rated on 5 now let me and here thank you we have a but the that thank you for your period presentation any questions
and I get to question the 1st 1 is that that is if you could affect the amount of water that that passes from 1 to the other by the amount of water that that the interest rate into the ground and the 2nd 1 is and it is like
yes not by the end of time and it's about the competition aspect auditing at any iteration directory not really complicated get man in a way funding and that get rid of that no right and when you have
that then you would have the hydrologic model with
infiltration and all the things like that OK and during the 2nd 1 is why do keep programs when they homework because there maddening males and I have a lot be aware what model people of
applications that I and you have that many people are working with people like that but the number of people really get that by they pass is increasing we are trying to be
ahead of time that when the 2 you you should we move from words
group so we are working with others that will read that you only you're going to have a good relationship but without the both the exchange between this going to work in the rain that we deal because we don't know the bulk of the of the book of the year you repeat work because I that we not later part content to the thing and that the few
words fell on the carrier attention diagram that the exactly that mainly handle all the they found by the knowledge transfer between main memory and this is done by the algorithm that the virtual memory numbering it and are that something that you can do it by completely redesigning gallery 0 yeah it can be done 2 you the shoemaking further because you couldn't you you know and environment got an image of a lot of no because of the computers in the middle of the of the of the technology to do so I probably would people being put in what is the effect of commitment the bulk of the book but you will do you the move both the but because it would right of the power factor that our method that the main thing is that have the title of Emperor and Bayesian look at it and don't from those that finally the other entire optimal and we that we need to get back to the date that not so important in but don't know
the extent the embedded the the the the 18 you look at the decoding occurred in data but we think it will look at and I don't have time for that and I out the data that I have and on and on and on actually all of the data are in that we want the data or we have them from people and on the environment and you note if there are like that is being that be the input output the efficient algorithm can be used also for other during the modules in graphs and the 2nd question is so that group think here a extracted in the future improvement over your program and extract extract the volume model line from our combination might be at the core are but watershed it also produces the online so you can extract like are you referring to limit the word of the the 1st question why it can be applied to other problems but is that we can only hope so and that they on on the conclusion Future work life I here you know if any of the mind and the more I Ireland applications then as it did a happy to hear about that 80 other question why did fraction of the value and by I don't know too much about that so I can't attend the you know how how it worried than I can tell think of an algorithm thank you I have to ask to stop here have the discussion because the
Besprechung/Interview
Bit
Punkt
Hochdruck
Versionsverwaltung
Gruppenkeim
Kartesische Koordinaten
Kardinalzahl
Computeranimation
Gradient
Videokonferenz
Richtung
Metadaten
Algorithmus
Einheit <Mathematik>
Font
Mustersprache
Meter
Maschinelles Sehen
Quellencodierung
Regulator <Mathematik>
Funktion <Mathematik>
Bildauflösung
Nichtlinearer Operator
Lineares Funktional
Addition
Parametersystem
Datennetz
Kategorie <Mathematik>
Template
Güte der Anpassung
Abfrage
Ideal <Mathematik>
Plot <Graphische Darstellung>
p-Block
Biprodukt
Teilbarkeit
Konfiguration <Informatik>
Rechenschieber
Druckverlauf
Rechter Winkel
Digitalisierer
Festspeicher
Information
Ordnung <Mathematik>
Programmierumgebung
Lesen <Datenverarbeitung>
Instantiierung
Telekommunikation
Gewicht <Mathematik>
Ortsoperator
Wasserdampftafel
Interaktives Fernsehen
Zellularer Automat
Term
Mathematische Logik
Physikalische Theorie
Virtuelle Maschine
Spannweite <Stochastik>
Domain-Name
Software
Reelle Zahl
Globale Optimierung
Datentyp
Programmbibliothek
Optimierung
Drei
Informatik
Leistung <Physik>
Videospiel
Graph
Eingebettetes System
Gibbs-Verteilung
Binder <Informatik>
Datenfluss
Quick-Sort
Energiedichte
Flächeninhalt
Mereologie
Simulation
Wort <Informatik>
Wärmeausdehnung
Größenordnung
Verkehrsinformation
Gewicht <Mathematik>
Rahmenproblem
Zahlenbereich
Strömungsrichtung
p-Block
Element <Mathematik>
Videokonferenz
Gradient
Algorithmus
Flächeninhalt
Festspeicher
Mini-Disc
Vorlesung/Konferenz
Ordnung <Mathematik>
Orakel <Informatik>
Sichtbarkeitsverfahren
Punkt
Wasserdampftafel
Abgeschlossene Menge
Zahlenbereich
Element <Mathematik>
Computerunterstütztes Verfahren
Räumliche Anordnung
Komplex <Algebra>
Physikalische Theorie
Computeranimation
Richtung
Bildschirmmaske
Informationsmodellierung
Algorithmus
Font
Netzbetriebssystem
Minimum
Mustersprache
Datentyp
Bildschirmfenster
Ganze Funktion
DoS-Attacke
Soundverarbeitung
Softwaretest
Befehl <Informatik>
Graph
Routing
p-Block
Ein-Ausgabe
Datenfluss
Teilbarkeit
Unendlichkeit
Konfiguration <Informatik>
Summengleichung
Rechenschieber
Flächeninhalt
Festspeicher
Mereologie
Schwimmkörper
Projektive Ebene
Faktor <Algebra>
Ordnung <Mathematik>
Punkt
Wasserdampftafel
Zahlenbereich
Kombinatorische Gruppentheorie
Richtung
Data Mining
Virtuelle Maschine
Informationsmodellierung
Radikal <Mathematik>
Vorlesung/Konferenz
Parallele Schnittstelle
Leistung <Physik>
Funktion <Mathematik>
Metropolitan area network
Lineares Funktional
Parametersystem
Schwellwertverfahren
Datennetz
Green-Funktion
Plot <Graphische Darstellung>
Ein-Ausgabe
Frequenz
Datenfluss
Konfiguration <Informatik>
Kollaboration <Informatik>
Flächeninhalt
Automatische Indexierung
Rechter Winkel
Festspeicher
Mereologie
Wort <Informatik>
Programmierumgebung
Fitnessfunktion
Bell and Howell
Wasserdampftafel
Besprechung/Interview
Bitrate
Rechter Winkel
Besprechung/Interview
Iteration
Verzeichnisdienst
Metropolitan area network
Informationsmodellierung
Zahlenbereich
Vorlesung/Konferenz
Kartesische Koordinaten
Optimierung
Hyperbelverfahren
Mereologie
Gruppenkeim
Wort <Informatik>
Inhalt <Mathematik>
Schaltnetz
Besprechung/Interview
Gruppenkeim
Wärmeübergang
Kartesische Koordinaten
Computerunterstütztes Verfahren
Ungerichteter Graph
Hinterlegungsverfahren <Kryptologie>
Informationsmodellierung
Algorithmus
Spezifisches Volumen
Maßerweiterung
Virtuelle Adresse
Bildgebendes Verfahren
Gerade
Funktion <Mathematik>
Soundverarbeitung
Videospiel
Bruchrechnung
Freier Ladungsträger
Ein-Ausgabe
Modul
Diagramm
Festspeicher
Wort <Informatik>
Speicherabzug
Decodierung
Programmierumgebung

Metadaten

Formale Metadaten

Titel Flow computation on massive grid terrains
Serientitel Open source GIS - GRASS user conference 2002
Anzahl der Teile 45
Autor Toma, Laura
Arge, Lars
Mitasova, Helena
Lizenz CC-Namensnennung - keine Bearbeitung 3.0 Deutschland:
Sie dürfen das Werk in unveränderter Form zu jedem legalen Zweck nutzen, vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen.
DOI 10.5446/21759
Herausgeber University of Trento
Erscheinungsjahr 2002
Sprache Englisch

Technische Metadaten

Dauer 28:30

Inhaltliche Metadaten

Fachgebiet Informatik

Ähnliche Filme

Loading...