mobile grid

Design of high performance networking platform considering mobile environment Tai M. Chung Internet Management Technology Laboratory School of Electrical and Computer Engineering SungKyunKwan University 300 Chunchun-dong, Changan-gu, Suwon, Kyounggi-do, 440-746, Korea Tel: +82-31-290-7131, Fax: +82-31-299-6673 1 Contents Introduction & Concept Scheduling algorithm Implementations Conclusions & Future Works 2 Introduction & Concepts 3 Introduction  Convergence of mobile and Grid system  Extend the Grid’s resource model to wireless mobile network  Allow access to Grid system using the ubiquitous mobile devices  Information gathering using wireless sensor networks  Why wireless?  Explosive increase of Mobile Device  New pervasive technology CPU Speed, Storage Capacity Coverage: Ubiquity Bandwidth: very high-speed Mobility management: Mobile IP, Ad-hoc routing 4 Mobile Grid Wired Grid Wireless Grid Mobile Node Access Point Mobile Node Proxy Access Point Mobile Node  Powerful machines with large amounts of memory and very fast processors  High-bandwidth links, disconnections are due to either explicitly performed or failures  Static location, hosts can be added, deleted or moved  Limited capabilities: slow CPU speed, little memory, low battery power and small screen size  Unpredictable disconnections is considered as a part of normal wireless communication  Hosts may come and leave generally mu ch more rapidly  Bandwidth and quality of the network connection may vary greatly 5 Problem Statement  Frequent changes in the environments  Variability of network bandwidth, mobility  Temporary and unannounced loss of network connectivity  Resource Heterogeneity  Ranging from powerful laptops to human WAP-enabled cell phones  Scarce resources  slow CPU speed, little memory, low battery power  Variety of user interface  Variety of screen size, input scheme 6 Research Objective  To suggest the mobile grid architecture considering wireless network  To identify and formulate the response time function in mobile grid  To determine the scheduling algorithm of task allocation using the response time function  To establish the mobile grid testbed with the PDAs in the wireless LAN network environment  To show the efficiency of mobile grid architecture using the mathematical analysis and simulation 7 System Architecture for Mobile Grid  Challenge: Integrating Mobile Wireless Devices into the Computational Grid  Thomas Phan, Lloyd Huang, and Chris Dulan  Communication Paradigm for Mobile Grid Users  Dario Bruneo, Marco Scarpa, Angelo Zaia, and Antonio Puliafito  Extending the Condor Distributed Systems for Mobile Clients  Song-Yi Yi and Miron Livny  SETI@Anywhere  Agile Computing 8 Scheduling Algorithm 9 MG-JSA Model  MG-JSA Model: Mobile Grid Job Scheduling Algorithm  Prediction of task processing time  Predicting the response time of task processing using the mathematical model  Task allocation and scheduling algorithm  Partitioning the parallel tasks and allocate the subtasks to appropriate mobile nodes to achieve efficient performance  Task processing monitoring until task completed  Rescheduling the task allocation process considering the response time when unexpected events are occurred 10 Process Transition of MG-JSA Resource monitoring of mobile nodes resource information gathering retrieve retrieve Prediction of task processing time retrieve Scheduling algorithm of task allocation distribute the Job in wireless network Rescheduling algorithm subtask fail Reallocation the subtask Partition and distribution the job Task processing monitoring all subtask are completed Task completion 11 Task allocation and scheduling algorithm  Job J arrives in proxy server  proxy server checks the available mobile nodes  collect the available resources information of mobile nodes  for (k = 1 ; k < m ; k ++)  partition and allocate subtasks to each workstation using the prediction of task processing time  monitoring the job processing until job completed  if subtaskk fails then reallocate the failed subtaskk to other available mobile nodes considering the early finish time  if unexpected event is occurred at mobile node k then evaluate the processing time between the original scheduling and rescheduling  if original scheduling (time) > rescheduling (time) then rescheduling  compute the T until Tk > Tk+1  The results of Job J returned 12 Prediction of task processing time  Prediction of transfer time in wireless network α Pc Connection State Mobile Grid Nodes Pd Disconnection State Mobile Grid Nodes β [State transition diagram of mobile node]  Probability of connection state : Pc  Probability of disconnection state : Pd  Disconnection rate : α  Reconnection : β  Then, the prediction time of data transfer for t time unit data is like this : 1  f (t )  t  ( )t Pc  13 Prediction of task processing time  Prediction of processing time in non-dedicated system  Modeling parameters are like these :  W: total demand processing time  wk: demand of the parallel subtask processing time on mobile node k  m: number of available mobile nodes  S: the number of interruptions encountered  λk: rate of the job arrival poisson distribution at mobile node k  μk: sequential job service rate at mobile node k  ρk: utilization rate at mobile node k  Tk: parallel task completion time on mobile node k  W can be written like this W   wk k 1 m 14 Prediction of task processing time  Tk can be written like this : Tk  X1  Y1  X 2  Y2    X S  YS  Z Local job Subtask X1 Y1 X2 Y2 …… Tk …… XS YS Z Xi : computing time consumed by the parallel task Yi : computing time consumed by the sequential jobs Z : execution time of the last parallel process that finishes the parallel task  Assumption for modeling  Owner job arrival process follows a Poisson distribution  Xi is an exponentially distributed random variable  Owner job processing follows M/G/1 15 Prediction of task processing time  Then we can acquired the results like these:  Mean of Tk E (Tk )  1 wk 1  k  Generalization factor 1 m  CPUspeed k m k 1  The mean of subtask completion time at mobile node k ck  CPUspeed k Tk  W (mc k    k ) k 1 m 16 Performance Evaluation  Mathematical Analysis  Network environments are suggested in the paper of S. Radhakrishnan et al  Stable (α: 0.003 ~ 0.001, β: 0.003~0.001)  Highly disconnective (α: 0.027 ~ 0.009, β: 0.003~0.001)  Unstable (α: 0.027 ~ 0.009, β: 0.027 ~ 0.009)  Highly connective (α: 0.003 ~ 0.001, β: 0.027 ~ 0.009)  Parameters used in the mathematical analysis  Total required time to process the job(W) : 2000 sec  The number of available mobile node (m) : 1 ~ 10 nodes  time required for transferring input data (tin) : 1 sec  time required for transferring output data : 1 sec  ρand CPU speed of mobile nodes  Node1(0.2, 400Mhz), Node2(0.4, 600Mhz), Node3(0.3, 400Mhz)  Node4(0.4, 600Mhz), Node5(0.3, 400Mhz), Node6(0.2, 600Mhz)  Node7(0.6, 400Mhz), Node8(0.4, 600Mhz), Node9(0.5, 400Mhz)  Node10(0.7, 600Mhz) 17 Performance Evaluation Stable 900 800 700 600 500 400 300 200 100 0 1 3 5 7 number of mobile nodes 9 900 800 700 600 500 400 300 200 100 0 1 3 5 7 number of mobile nodes 9 Highly disconnective time (sec) time (sec) MG-JSA AppLeS MG-JSA AppLeS Unstable 900 800 700 600 500 400 300 200 100 0 1 3 5 7 number of mobile nodes 9 Highly connective 900 800 700 600 500 400 300 200 100 0 1 3 5 7 number of mobile nodes 9 time (sec) time (sec) MG-JSA AppLeS MG-JSA AppLeS 18 Implementations 19 MGA & MGP  Mobile Grid Architectures & Mobile Grid Protocol  Requirements  Dynamic Resource Management  Active and efficient discovery & management of resource and service of mobile node ,considering wireless network  Quality of Service Guarantee  Reliable request/distribution of job to mobile node, accomplishment of jobs, collection of results  Security  User authentication and encrypted messages because wireless network is weak than wired network  Interoperability  Interoperation between wired-GRID 20 Design of MGA  Proxy Part Resource Manager User DB Job Service Manager Resource Information DB Job Information DB Job Management Job Scheduler User Authenticator Survivability Manager Communicator Job Process Parallel Process Management M O B I L E N O D E 21 Job Information Survivability Check Resource Information Design of MGA  Mobile node Part User Interface Super Daemon U S E R Operation Results Display Resource Information Display Job Execution Module System State Management Module Resource Manager MGP module Node Controller P R O X Y System Kernel Processor Memory Power 22 Design of MGP  Messages  Login & initialization message  Periodic resource information message  Job request/reply message  Survivability check message  Special features  Format  Header + Payload  Nested TLV ID Rbit Version Total_Length Flag Header Payload  Periodic resource request  For job scheduling algorithm  For decrease operation delay Type Type Length Value Length Value Type Length Value 23 Operations Rep_login_Accept : Authentications Rep_login_Accept Req_outArea Rep_outArea Rep_outArea : mobile node replies when it can communicate with proxy Req_Login Req_Login : insert user ID/Password and initialization Join to Grid network Req_outArea : Check abnormal communication or disconnection Rep_outArea : Estimates that mobile node is capable Req_Resource : Broadcast the message for getting resource information Req_Resource Rep_Resource Rep_Resource : receive the resource information of mobile node and save in DB Job Scheduling Rep_Resource : reply information of the mobile to proxy through Resource Management Module Req_Job Req_Job : user can request the job Parallel processing through MPICH Rep_Job : Reply the job results MPICH Module Rep_Job Mobile node reply the results of distributed jobs to proxy after parallel processing Initialization Notify_Out_Area Notify_Out_Area : If mobile node want to be out of grid, this message 24 is sent Results of Implementation Proxy Mobile node 25 Conclusions & Future Works 26 Conclusions  Design and implementation of the mobile grid system  MGA(Mobile Grid Architecture)  MGP(Mobile Grid Protocol)  Mobile grid testbed using PDAs in wireless LAN networks  MG-JSA (Mobile Grid Job Scheduling Algorithm)  Giving the useful information of job partition  Selection of the appropriate mobile node and it’s number  Rescheduling the job allocation in time of job failure  Performance Evaluation  Mathematical analysis  Simulation in the testbed 27 Future Works  Extend the mobile grid system considering the various mobile nodes and other network environments  Inclusion the security mechanisms for mobile communication  Suggesting the useful applications which can be used in mobile grid network  Suggesting the business model of mobile grid network 28