La présentation est en train de télécharger. S'il vous plaît, attendez

La présentation est en train de télécharger. S'il vous plaît, attendez

Eddy Caron - ACI GDS – 22/09/03 1 ACI MD GDS Le middleware pour GDS

Présentations similaires


Présentation au sujet: "Eddy Caron - ACI GDS – 22/09/03 1 ACI MD GDS Le middleware pour GDS"— Transcription de la présentation:

1 Eddy Caron - ACI GDS – 22/09/03 1 ACI MD GDS Le middleware pour GDS http://graal.ens-lyon.fr/~diet

2 Eddy Caron - ACI GDS – 22/09/03 2 Plan Réservation de ressource dans un ASP hiérarchique Déploiement automatique DIET en P2P DIET vs NetSolve VizDIET Communications dans DIET Une application pour GDS

3 Eddy Caron - ACI GDS – 22/09/03 3 Context One long term idea for Grid computing: renting computational power and memory capacity over the Internet – Very high potential Need of Problem Solving Environments (PSEs) Applications need more and more memory capacity and computational power Some proprietary libraries or environments need to stay in place Difficulty of installation for some libraries or applications Some confidential data must not circulate over the net Use of computational servers accessible through a simple interface Need of schedulers Moreover … –Still difficult to use for non-specialists Almost no transparency Security and accounting issues usually not addressed –Often application dependent PSEs –Lack of standards (CORBA, JAVA/JINI, sockets, …) to build the computational servers

4 Eddy Caron - ACI GDS – 22/09/03 4 RPC and Grid-computing  GridRPC A simple idea –RPC programming model for the Grid –Use of distributed collections of heterogeneous platforms on the Internet –For applications require memory capacity and/or computational power –Task parallelism programming model (synchronous/asynchronous) + data parallelism on servers  mixed parallelism Needed functionality –Load balancing resource discovery performance evaluation Scheduling –Fault tolerance, –Data redistribution, –Security, –Interoperability, …

5 Eddy Caron - ACI GDS – 22/09/03 5 GridRPC AGENT(s) S1 S2 S3S4 A, B, C Answer (C) S2 ! Request Op(C, A, B) Client

6 Eddy Caron - ACI GDS – 22/09/03 6 GridRPC (con’t) 5 main components: –Client submits problems to servers Gives users’ interfaces –Server solves problems sent by clients Runs software –Database contains dynamic and static information about software and hardware resources –Scheduler chooses an appropriate server depending of –the problem sent –the information contained in the database –Monitor gets information about the status of the computational resources

7 Eddy Caron - ACI GDS – 22/09/03 7 DIET - Distributed Interactive Engineering Toolbox - Hierarchical architecture for an improved scalability Distributed information in the tree Plug-in schedulers MA A LA Direct connection Server front end Master Agent

8 Eddy Caron - ACI GDS – 22/09/03 8 A B C FAST - Fast Agent’s System Timer - NWS-based (Network Weather Service, UCSB) Computational performance –Load, memory capacity, and performance of batch queues (dynamic) –Benchmarks and modeling of available libraries (static) Communication performance –To be able to guess the data redistribution cost between two servers (or between clients and servers) as a function of the network architecture and dynamic information –Bandwidth and latency (hierarchical)

9 Eddy Caron - ACI GDS – 22/09/03 9 PIF - Propagate Information Feedback - Algorithm from distributed system research –PIF : Propagate Information Feedback Two steps –First phase: broadcast phase Broadcast one message through the tree –Second phase: feedback phase When the leaf has no descendant  feedback message is sent to their parent When the parent receives the feedback messages from all its descendants, it sends a feedback message to its own parent, and so on

10 Eddy Caron - ACI GDS – 22/09/03 10 PIF and DIET - broadcast phase - AA LA S1S2S3S4S5S6S7S8S9S10S11S12S13S14S15S16 1. Broadcast the client’s request 2. Sequential FAST interrogation for each LA 3. Resource reservation MA

11 Eddy Caron - ACI GDS – 22/09/03 11 PIF and DIET - feedback phase - AA LA S1S2S3S4S5S6S7S8S9S10S11S12S13S14S15S16 1. chooses the identity of the most (or list of) ``appropriate'' server(s) 2. unused resources are released S4S7 S12S15 MA

12 Eddy Caron - ACI GDS – 22/09/03 12 PIF and DIET - feedback phase - AA LA S1S2S3S4S5S6S7S8S9S10S11S12S13S14S15S16 1. chooses the identity of the most (or list of) ``appropriate'' server(s) 2. unused resources are released S4 S12 MA

13 Eddy Caron - ACI GDS – 22/09/03 13 PIF and DIET - feedback phase - AA LA S1S2S3S4S5S6S7S8S9S10S11S12S13S14S15S16 1. chooses the identity of the most (or list of) ``appropriate'' server(s) 2. unused resources are released S12 MA

14 Eddy Caron - ACI GDS – 22/09/03 14 Server failure and reactivity Take into account server failure and increase the DIET reactivity Time out at the LA level : –Dead Line 1 = ß1* Call_FAST_time + ß2 * nb_server AA LA S1S2S3S4S5S6S7S8S9S10S11S12S13S14S15S16 MA S7 S12S15 S2 [DEAD LINE 1] S2

15 Eddy Caron - ACI GDS – 22/09/03 15 Hierarchical fault tolerance No answer after dead line 1 –Dead Line 2 = ß3* level_tree AA LA S1S2S3S4S5S6S7S8S9S10S11S12S13S14S15S16 MA S7 S12S15 S7 [DEAD LINE 2]S12

16 Eddy Caron - ACI GDS – 22/09/03 16 Simulation: SimGRID2 Real experiments or simulations are often used to test or to compare heuristics. Designed for distributed heterogeneous platforms Simulations can enable reproducible scenarios Simgrid: a distributed application simulator for scheduling algorithm evaluation purposes. –Event-driven simulation –Fixed or According to a trace values to characterize SimGrid resources Processors Network links Simgrid2 : A simulator built using SG. This layer implements realistic simulations based on the foundational SG and is more application- oriented. –Simulations are built in terms of communicating agents.

17 Eddy Caron - ACI GDS – 22/09/03 17 The DIET SimGRID2 simulator

18 Eddy Caron - ACI GDS – 22/09/03 18 Evaluation of the PIF scheduler

19 Eddy Caron - ACI GDS – 22/09/03 19 Conclusion and future work Conclusion –Benefit from distributed system research –Fault tolerance into DIET Server failure Branch failure –Resource reservation performs a good QoS for client requests –DIET SimGRID2 simulator Can be reused to validate other algorithms Future work –Implementation of some tools to guarantee the resource reservation –Integrate NWS trace into the simulator –How to fix deadlines on a given heterogeneous platform ?

20 Eddy Caron - ACI GDS – 22/09/03 20 Plan Réservation de ressource dans un ASP hiérarchique Déploiement automatique DIET en P2P DIET vs NetSolve VizDIET Communications dans DIET Une application pour GDS

21 Eddy Caron - ACI GDS – 22/09/03 21 Automatic deployment Problem : Take the right number of components (resources) and place them in right way, to increase the overall performance of the platform. Motivation : “how to deploy DIET on the grid ?” Foundation : Idea given in the article “Scheduling strategies for master- slave tasking on heterogeneous processor grids” by C.Banino, O.Beaumont, A. Legrand and Y.Robert.

22 Eddy Caron - ACI GDS – 22/09/03 22 Introduction Solution : –Generate a new structure by arranging the resources according to the graph, which gives best throughput. –For homogeneous platform, the resources should be arranged in a binary tree type structure. –For heterogeneous platform, more resources should be added by checking the bottleneck in the structure.

23 Eddy Caron - ACI GDS – 22/09/03 23  Architectural model - Deployment

24 Eddy Caron - ACI GDS – 22/09/03 24  w i (Mflp/s): Computing power of node P i  bij : capacite of link (links are symmetric and bidirectional)  S in i : size of incomming request from client  S out i : size of outgoing request (response)  alpha in i : fraction of time for the computation of incomming request  alpha out i : fraction of time taken for the computaion of outgoing resuest Deployment

25 Eddy Caron - ACI GDS – 22/09/03 25 Calculation of the throughput of a node PiPi PjPj b ij Operations in steady state

26 Eddy Caron - ACI GDS – 22/09/03 26 Calculation of the throughput of a graph

27 Eddy Caron - ACI GDS – 22/09/03 27 Min(9,3+4)Min(7,6+5) Min(12,7+7) Min(20,12)20 5 4 3 6 79 12 7 Example: Calculation of throughput of graph

28 Eddy Caron - ACI GDS – 22/09/03 28 All nodes have same computing power and bandwidth link Binary Tree Chain Graph Star Graph 2 Chain Graph 2 Depth Star Graph Homogeneous Structures

29 Eddy Caron - ACI GDS – 22/09/03 29  Simulation results (with 8nodes) - Homogeneous Structures

30 Eddy Caron - ACI GDS – 22/09/03 30  Simulation results (with 32 nodes) - Homogeneous Structures

31 Eddy Caron - ACI GDS – 22/09/03 31  Simulation results (for Binary graph) - Homogeneous Structures

32 Eddy Caron - ACI GDS – 22/09/03 32 Heterogeneous Networks

33 Eddy Caron - ACI GDS – 22/09/03 33 1  40 1  15 1  200 R = 2 24 17.88 19 15 29 0.052 0.88 0.31 30 25 30 Throughput of network

34 Eddy Caron - ACI GDS – 22/09/03 34 0.11 1.05 1  200 1  40 1  15 24 17.88 19 15 29 0.052 0.88 0.1167 R = 2R = 2.2 R = 2.65 1  100 1  20 Throughput of network by adding LAs

35 Eddy Caron - ACI GDS – 22/09/03 35 Heterogeneous Network

36 Eddy Caron - ACI GDS – 22/09/03 36 Experimental results 1 client with n requests –no steady state (MA performed good) n clients with n requests –no steady state (MA performed good) –Pipeline effect –not enough nodes (clients) Buffered the requests at MA –a new client implementation to make an effect of steady state –MA failed with 960 requests (due to memory problem)

37 Eddy Caron - ACI GDS – 22/09/03 37 Experimental results

38 Eddy Caron - ACI GDS – 22/09/03 38  Select best structure  Improve the throughput of the network  Predict the performance of the structure  Can find the effects on performance if different changes are done in the structure configuration  Bottleneck is not caused at MA Conclusion

39 Eddy Caron - ACI GDS – 22/09/03 39  Homogeneous :  Binary tree type structure is best  Number of nodes is proportionate to number of servers  Star graph type structure, when nodes are less and servers are more than 60  Heterogeneous :  Find the bottleneck  Improve the throughput  Modelizing the DIET Conclusion

40 Eddy Caron - ACI GDS – 22/09/03 40  Calculate the throughput of structures with multi-client and multi-master agents.  Dynamic updating with the use of package GRAS  Timer addition into the tool to get real value for CORBA implementation of DIET  Check the LA and SeD as the cause of bottleneck  Combine scheduling and deployment to increase the performance  Validation of work by real deployment. Future work

41 Eddy Caron - ACI GDS – 22/09/03 41 193.253.175.223 /home/ckochhof/work/diet/src/dmat_manips/bin/ma1 193.253.175.224 /home/ckochhof/work/diet/src/dmat_manips/bin/LA1 193.253.175.226 /home/ckochhof/work/diet/src/dmat_manips/bin/SeD1 all 193.253.175.227 /home/ckochhof/work/diet/src/dmat_manips/bin/SeD2 T:MatSUM 193.253.175.225 /home/ckochhof/work/diet/src/dmat_manips/bin/LA2 193.253.175.228 /home/ckochhof/work/diet/src/dmat_manips/bin/SeD3 T:MatSUM #!/usr/bin/perl -w use strict; use XML::DOM; my $parser = XML::DOM::Parser->new(); my $file = ‘launch.xml'; my $doc = $parser->parsefile ($file); package MasterAgent; sub MasterAgent::new { my $class = shift; my $self = {};$self->{type} = undef;$self->{name}= undef;$self->{IP}= undef; $self->{localAgents}= {};$self->{server} = {};bless $self,$class;$self; } sub MasterAgent::funtion{ my $self = shift; my $agt = shift; $self->{name}= $agt->getAttribute('name');; $self->{type} = "MasterAgent"; my @child = $agt->getChildNodes; for(my $i=0;$i<$#child;$i++){ if ($child[$i]->getNodeName == 'IP'){ $self->{IP} = $child[$i]->getData;} if ($child[$i]->getNodeName == 'localagent'){ my $local_name = $child[$i]->getData; my $child_pass = $child[$i]; my $obj_la_ma = new LocalAgent; $self->{localAgents}->{$local_name} = $obj_la_ma->funtion($child_pass,$self);} if ($child[$i]->getNodeName == 'server'){ my $server_name =$child[$i]->getData; my $child_pass = $child[$i]; my $obj_ser_ma = new Server; $self->{servers}->{server_name}= $obj_ser_ma->funtion($child_pass,$self);} } return $self; } Master Agent :. Name = MA1. Binary = /home/ckochhof/work/diet/src/dmat_manips/bin/ma1. IP = 193.253.175.223. Local agent = LA1 LA2 Local Agent :. Name = LA1. Parent agent= MA1. Binary = /home/ckochhof/work/diet/src/dmat_manips/bin/LA1. IP = 193.253.175.224. servers = SeD1 SeD2 Server :. Name = SeD1. Parent agent = LA1. binary = /home/ckochhof/work/diet/src/dmat_manips/bin/SeD1. IP = 193.253.175.226. services = all Server :. Name = SeD2. Parent agent = LA1. Binary = /home/ckochhof/work/diet/src/dmat_manips/bin/SeD2. IP = 193.253.175.227. services = T:MatSUM Local Agent :. Name = LA2. Parent agent= MA1. Binary = /home/ckochhof/work/diet/src/dmat_manips/bin/LA2. IP = 193.253.175.225. servers = SeD3 Server :. Name = SeD3. Parent agent = LA2. Binary = /home/ckochhof/work/diet/src/dmat_manips/bin/SeD3. IP = 193.253.175.228. services = T:MatSUM traceLevel = 1 agentType = DIET_LOCAL_AGENT name = LA1 Parent name = MA1 fastUse = 1 ldapUse = 0 nwsUse = 1 LA1.cfg Script to launch the agents. Simulations to check the parser Practical implementation To Do Automatic Deployment : first tool

42 Eddy Caron - ACI GDS – 22/09/03 42 Plan Réservation de ressource dans un ASP hiérarchique Déploiement automatique DIET en P2P DIET vs NetSolve VizDIET Communications dans DIET Une application pour GDS

43 Eddy Caron - ACI GDS – 22/09/03 43 DIET en P2P Existant –Multi-MA disponible avec connection en JXTA –Docs disponibles –Archive disponible : diet-0.7_beta-dev-jxta.tgz TODO list –Evaluer les performances –Vérifier le respect des coding standard –Intégration au CVS DIET –Briser la contrainte 1 composant JXTA pour 1 composant DIET –Algorithmes “intelligents” pour le parcours des MA ? MA A LA Connexions JXTA

44 Eddy Caron - ACI GDS – 22/09/03 44 Plan Réservation de ressource dans un ASP hiérarchique Déploiement automatique DIET en P2P DIET vs NetSolve VizDIET Communications dans DIET Une application pour GDS

45 Eddy Caron - ACI GDS – 22/09/03 45 DIET vs NetSolve Scripts de déploiement. Utilisation de CVS pour mettre à jour les fichiers de configuration. paraski ls sunlabs gateway_vthd clients agents + servers

46 Eddy Caron - ACI GDS – 22/09/03 46 DIET vs NetSolve

47 Eddy Caron - ACI GDS – 22/09/03 47 DIET vs NetSolve TODO List –Tests avec API asynchrone. –'Multithreader' le client. –Amélioration des statistiques (indice de dispersion). –Amélioration des scripts de déploiements (fichiers de configuration DIET et omniORB). –Expliquer les résultats de NetSolve –Expliquer le problème des 40 clients de DIET –Tests sur les SPARC –Tests icluster2 ?

48 Eddy Caron - ACI GDS – 22/09/03 48 Plan Réservation de ressource dans un ASP hiérarchique Déploiement automatique DIET en P2P DIET vs NetSolve VizDIET Communications dans DIET Une application pour GDS

49 Eddy Caron - ACI GDS – 22/09/03 49 VizDIET Chaque LogManager collecte les infos de son agent et les envoie au LogCentral situé en dehors de la structure DIET. VizDiet : –outil de visualisation en Java –Interraction sur la plate-forme

50 Eddy Caron - ACI GDS – 22/09/03 50 VizDIET 1.0 Intégration de LogService (LogManager/LogCentral) dans les agents DIET Transfert de messages depuis l'agent par l'intermédiaire du LogManager –pas de stockage sur disque Etude vizPerf vs vizDIET –Conclusion : vizPerf trop éloigné de la structure DIET

51 Eddy Caron - ACI GDS – 22/09/03 51 VizDIET 2.0 VizDiet collectera les infos de la structure DIET par LogCentral et les affichera en temps réel. VizDiet doit pouvoir aussi agir sur la structure en générant des scripts XML (flèche rouge) ou en modifiant des infos XML existantes (flèche bleue)

52 Eddy Caron - ACI GDS – 22/09/03 52 Screenshot

53 Eddy Caron - ACI GDS – 22/09/03 53 Plan Réservation de ressource dans un ASP hiérarchique Déploiement automatique DIET en P2P DIET vs NetSolve VizDIET Communications dans DIET Une application pour GDS

54 Eddy Caron - ACI GDS – 22/09/03 54 Du neuf dans les communications DIET Asynchrone : –Finaliser la compatibilité GridRPC (erreur, handle,…) –Bug sur le diet_wait_and ? –Fuites mémoires ? (test d’échelle) PadicoTM –Compilation/Test –Intégration dans DIET en cours –Limitation des plate-formes utilisables –Bug sur le dechargement des modules (lié à la fonction ‘dlopen’ selon la ‘libc’ utilisée)

55 Eddy Caron - ACI GDS – 22/09/03 55 Plan Réservation de ressource dans un ASP hiérarchique Déploiement automatique DIET en P2P DIET vs NetSolve VizDIET Communications dans DIET Une application pour GDS

56 Eddy Caron - ACI GDS – 22/09/03 56 Une application pour GDS ? GriPPS : Grid Protein Pattern Scanning –Application de bio-informatique –Pattern scanning Caractéristiques –Tâches répétitives, courtes mais très nombreuses –Entrées/sorties fichiers « texte » De quelques Mos à plusieurs Gos –Demande d’un temps de réponse faible

57 Eddy Caron - ACI GDS – 22/09/03 57 FAST


Télécharger ppt "Eddy Caron - ACI GDS – 22/09/03 1 ACI MD GDS Le middleware pour GDS"

Présentations similaires


Annonces Google