Merken

Parallel Inception

Zitierlink des Filmsegments
Embed Code

Automatisierte Medienanalyse

Beta
Erkannte Entitäten
Sprachtranskript
With the noise has not from the the like is held on wall will be on the long the views of the universe and that all the reasons are we going to do this so the cost of the which was which was the visionary who they PC books it's great science and the kind of I like the rest of the cost of belonging a little and I data with 2 means that decided to drive home the there are old enough of having like the the researchers mostly just what problems and the speed of you who brought you way that the more recently all of them but the traffic from the family structure you
were possible antecedents in the range of the but the role of his new things and this is not fair we still have a long time for the idea of what I'm in about so in in I should say this rule have more different things so I see more would be seeking warehousing data warehousing workloads which is pretty preacher on what I would assure you that is more on the rigid side not not reduced the so sequel isn't so so the most expressive way to describe the property in my in my mind for some people it's perfectly clear non-binders as constituting were you would you just can't cope with it and you can use these are used more often than it although I think it's it's the up it so it's not it's not quite as far as something that is fairly rigid they have thousands of course there was a loss that have described the things in the data is maximum so this is 1 of the things we can see so that's the kind of policies are intersecting and what I'm describing a framework for writing all this talk and there are some other things out there that that can bind to a more of these things so many people which I have looked at any rate I understand its components you wanted of the year and PG strong which is 1 of the rest of the technologies you used use the Internet joins so interesting things that we all know what closely and then there were of this sort is number which is from New analysis that's the kind of distribution if you basically is the Python code to be the vectorized in this work for you you were part of so very very cool library of I would recommend the light on the and the check up I put it in the context of this talk I kind of frame in this way of doing things you know and original varied kernel and a lot of this is how you have it to the user so as opposed to a group of students in his or her no source they're they're very well and allow status and see how far and really didn't do a very simple things very wide but can you like you boring on of course was change to and so but these things kind ought objects space we had really good sale on 1 side but on the other side we have really we have the capability to more complex things so the nice we can merge these in in 1 system published of
interest that so on this side of this is where you that's what you basically code and that code actually has a mother who the runs through this matter what is magical MySpace showing sales which is the pure language you can need the harder heterogeneity of themselves you used you used only you use use pretty flexible but not quite interesting that we would would call sort of but regardless described use very workflow you know it's less than revolvers you you have created the of contacts and the write program code which is actually in seeing this case number right you purify just and then you basically sites have to work with the special place so that's what this is here so this is actually the role that produce a very basic some of amazing tour certainly people have more than changing you writing the good things this but I I think the so
that's that's 1 half of the picture here have
this is where it becomes a so couple of examples that did this is what I want to the recently of source of the database which is offered by the ability of the of the source of the 1 can help you don't do anything the idea is the right 1 signals the year uh against the magic happens which really is that it's all sort of see but it is the master node that students that worked out to each of these individual worker nodes which is where we're going I would not work on this point that they put same idea you know each 1 of these is the sum that and executing network locally and that comes back as a single results so scatter gather carried out in very similar to what you would and now is the is the point where how we merge these 2 things right and so visited opposing if we take some some capabilities and you use that these cheese and he notes which have hundreds of a very low and of thousands of for the the price for there's been a very large the utility what was oratory larger you so
it will just step back a 2nd and think about how how that so where basically just close you right so I'll buy the that I code is dispatched to tens hundreds of thousands of course scalable will get but such a some number of nodes by libraries at the that who pilot and machine code execute machine code results come back and whole results might so that's a pretty powerful it's it's a not always use all of the source of most of of the world is the this is the other thing that you're taking some very large set in the national with the factor on how fast you can perform or and that's you know that's the name you know 1 you waiting around our statistical learning was all about it is the wish the role of a lot of time and so this is kind of you that this is a logical sort of logical so physical view of what a cluster complexity of the the piercing top some number of the nodes node has some kind of data it also has the green boxes which represent is due to the force of the 10 100 just has work but they is that they features each was sorry shot here is to further the understanding by 1 1 or more due course a but if you can see that in which you could do with all that kind of points so this all sounds great uh but what are the limitations right that's kind of the point you can just use things like this there is 1 of the honest about things that are about to which the BIO around this doesn't fit the problem all this is designed for the computation of the MLE tends to bring the data once and you travel times that's where you really the maximum of the watching as a notice durations additionally city about problems so if you have the flu and before the end of the next iteration is a function of the previous iteration then this doesn't work that way know that there's some problems where you can get around that but you know this is this is forcing the of and the last is rest so you may have many of the region you only has 1 can use it to annex constant for but you don't get to decide all users of queries on the same GPU units that have the but so all talk about a couple of use cases that the going too much time on this but I think this is 1 of our 1 of our larger datasets that we've seen part this tour I was thinking that this might be of so there's a you might you might ask a question like that how do I find that I have to say other measurements in this work for the agenda much the I wanna know something about how those of late I might do consultation is a lot of conditional doing things that so you might like that in this situation and this is my this method in a situation where each of these different types of the assessed the in parallel and this isn't that true positive over the others work in the home work but which is supported by you implementation of the year 2 years for Apple work experience working for the rich region of have another potential example which I think is is somewhat users have inspired from the around 2 the another part of the field but basically you saw of pictures here inside home automation others so called is you can imagine having your database on large numbers of features which usually don't think it is for that content but when you have this effect of and you can things like this so it's still very feature transform all of the the computer problems the light imagine to say how to say different things about so if you had say an image of an organism wanted know at which which prospects of something like 2 years later you some contracts with you and you can go up because my injury you may be able to put the subsystem to do that in a very interesting way you probably don't think of those problems about the last 1 is also really that data which currently is not that it's not that we support for geospatial rather there isn't that this work on the currently on but that's the being worked on 1 all time and it's also very difficult problems you can imagine that in the background you have a check that the files used for the style of a world cluster nodes and now you can write some kernel that is executing each styles the answer to it of this year lower cost so that I think the best way of finding out that at the end the so if you take cells from from all this I would say that again at stock is not actually making new names making new combinations of systems and things like that this is my experience is always the most difficult you bring your data application and you know what you interested in things that you actually have a large enough and that's the end of the delivery of application data relevant the universe the and that's working on this thing is that this is due to the use variable is a new with all the GCC is not going to get a very very projects which were downloaded not not anything that made the it's basically a modular have language you have some number of language versions of this year's because most of job so given all the work of force the the back of the hand abstracts in abstract representations program and on back and you have a particle assessment of it's the work scene or is here I wanted to give you the support that so it's a very powerful enough to get rid of this and last point which was out of myself and what is of a seamless and you will be right very powerful and saying the complexity of the problem stems from the fact that the the nice thing is that we have matrices positive requirements in this post so that you can handle that and that might be a power that you give them and things like that you can buy so that's this kind of talk I really get as many of the more interesting ones so that you can still see the idea of this you know the things that you want to watch and much of what we have to take questions probably has the from just crazy because anything so that's that's be if the some part of the solution that would not be that means that you know that you you you you did this so I am not familiar with a with how their how they are exactly the but I know that the use learning instantly and all of the of the 2 if this thing is the biggest thing so the source of much of the what the piece from is very interesting in the cell the competition so that the thing is the joints and the last 1 thing versions of and where do we
have
Arithmetisches Mittel
Sichtenkonzept
Pivot-Operation
Familie <Mathematik>
Geräusch
Supercomputer
Datenstruktur
Computeranimation
Wiederherstellung <Informatik>
Datenhaltung
Kernel <Informatik>
Distributionstheorie
Web Site
Einfügungsdämpfung
Data-Warehouse-Konzept
Rahmenproblem
Formale Sprache
Mathematisierung
MySpace
t-Test
Gruppenkeim
Schreiben <Datenverarbeitung>
Zahlenbereich
Fortsetzung <Mathematik>
Systemzusammenbruch
Programmcode
Code
Framework <Informatik>
Raum-Zeit
Kernel <Informatik>
Internetworking
Metropolitan area network
Spannweite <Stochastik>
Maßstab
Code
Programmbibliothek
Zusammenhängender Graph
Analysis
Schnelltaste
Kategorie <Mathematik>
Schlussregel
Physikalisches System
Quellcode
Kontextbezogenes System
Bitrate
Quick-Sort
Objekt <Kategorie>
Beanspruchung
Rechter Winkel
Mereologie
Resultante
Knotenmenge
Gewichtete Summe
Punkt
Datennetz
Rechter Winkel
Datenhaltung
Streuung
Softwarewerkzeug
t-Test
Einfache Genauigkeit
Quellcode
Resultante
Punkt
Extrempunkt
Selbstrepräsentation
Formale Sprache
Versionsverwaltung
Iteration
Kartesische Koordinaten
Computerunterstütztes Verfahren
Maschinensprache
Computer
Komplex <Algebra>
Kernel <Informatik>
Hoare-Logik
Freeware
Skalierbarkeit
Einheit <Mathematik>
Prozess <Informatik>
Parallele Schnittstelle
Einflussgröße
Lineares Funktional
Sichtenkonzept
Abstraktionsebene
Datenhaltung
Just-in-Time-Compiler
Abfrage
Quellcode
Teilbarkeit
Datenfeld
Forcing
Rechter Winkel
Datenparallelität
Projektive Ebene
Fitnessfunktion
Quader
Ortsoperator
Selbst organisierendes System
Multimedia
Schaltnetz
Zahlenbereich
Zellularer Automat
Implementierung
EDV-Beratung
Demoszene <Programmierung>
Knotenmenge
Variable
Datentyp
Programmbibliothek
Inverser Limes
Inhalt <Mathematik>
Optimierung
Algorithmische Lerntheorie
Grundraum
Bildgebendes Verfahren
Leistung <Physik>
Elektronische Publikation
Matrizenring
Zehn
Physikalisches System
Elektronische Publikation
Design by Contract
Mereologie
Partikelsystem
Speicherabzug
Computeranimation

Metadaten

Formale Metadaten

Titel Parallel Inception
Untertitel MPP Databases GPGPU
Serientitel FOSDEM 2016
Teil 32
Anzahl der Teile 110
Autor Dune, Kyle
Lizenz CC-Namensnennung 2.0 Belgien:
Sie dürfen das Werk bzw. den Inhalt zu jedem legalen Zweck nutzen, verändern und in unveränderter oder veränderter Form vervielfältigen, verbreiten und öffentlich zugänglich machen, sofern Sie den Namen des Autors/Rechteinhabers in der von ihm festgelegten Weise nennen.
DOI 10.5446/30942
Herausgeber FOSDEM VZW
Erscheinungsjahr 2016
Sprache Englisch

Inhaltliche Metadaten

Fachgebiet Informatik

Ähnliche Filme

Loading...