Does the implementation give solutions for the requirements? Flexibility Flexibility GridRPC enables dynamic join/leave of QM s ervers. GridRPC enables dynamic expansion of a QM server. Robustness Robustness GridRPC detects errors and application can implement a recovery code by itself. Efficiency Efficiency GridRPC can easily handle multiple cluster s. Local MPI provides high performance on a c luster by fine grain parallelism.
11
Embed
Does the implementation give solutions for the requirements? Flexibility GridRPC enables dynamic join/leave of QM servers. GridRPC enables dynamic expansion.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Does the implementation give solutions for the requirements?
FlexibilityFlexibilityGridRPC enables dynamic join/leave of QM servers.GridRPC enables dynamic expansion of a QM server.
RobustnessRobustnessGridRPC detects errors and application can implement a recovery code by itself.
EfficiencyEfficiencyGridRPC can easily handle multiple clusters.Local MPI provides high performance on a cluster by fine grain parallelism.
Strategy for long run
QM simulation will migrate to the other cluster either by intQM simulation will migrate to the other cluster either by intentionally or unintentionally.entionally or unintentionally.
intentional migrationExceeds the maximum runtime for the clusterReservation period has expired
unintentional migrationAny error/fault is detected
The next cluster will be selected by either reservation or siThe next cluster will be selected by either reservation or simple selection algorithm.mple selection algorithm.
Selection algorithm considersnumber of available cpusnumber of requested cpusrecords of past utilization
Simulation reads a host information file in every time step.Simulation reads a host information file in every time step.A cluster can join to/leave from the experiment on-the-fly.
National Institute of Advanced Industrial Science and Technology
Experiments- target simulation -
- testbed -- results and lessons learned -
Grid-enabled SIMOX Simulation on Japan-US Grid Testbed at SC2005
A technique to fabricate a micro A technique to fabricate a micro structure consisting of Si surface on structure consisting of Si surface on the thin SiOthe thin SiO22 insulator insulatorAllows to create higher speed with Allows to create higher speed with lower power consumption devicelower power consumption device
Oxygen implanted into Silicon substrate
SiO2 insulator
Annealing
IC IC
Creating IC chips on the Insulator
leak current
Formation of Silicon over insulator (SOI) structure
SIMOX simulation on the Grid
Simulate SIMOX by implanting five oxygen atoms with their initial velocities Simulate SIMOX by implanting five oxygen atoms with their initial velocities much smaller than the usual values.much smaller than the usual values.
The incident positions of the oxygen atoms relative to the surface crystalline The incident positions of the oxygen atoms relative to the surface crystalline structure of Si differ.structure of Si differ.
5 QM regions are initially defined5 QM regions are initially definedSize and No. of QM regions are changed during the simulation
0.11million atoms in total0.11million atoms in total
Results of the experiments will demonstrate the sensitivity of the process on Results of the experiments will demonstrate the sensitivity of the process on the incident position of the oxygen atom when its implantation velocity is the incident position of the oxygen atom when its implantation velocity is small.small.
Testbed for the experiment
Phase 1 Phase 2 Phase 3 Phase 4
AIST Super ClustersAIST Super ClustersP32 (2144 CPUs), M64 (528 CPUs), F32 (536 CPUs)
Results of the experiment (cont’d)Behavior of oxygen atoms strongly Behavior of oxygen atoms strongly depends on the incident positiondepends on the incident position
0
0.2
0.4
0.6
0.8
1
1 51 101 151 201 251
QM 1QM 2QM 3QM 4QM 5
v/v0
Time step
Expanding/Dividing QM regions at every 5 Expanding/Dividing QM regions at every 5 time stepstime steps (( Expansion: 47 times, Division: 8 times)Expansion: 47 times, Division: 8 times)
0100200300400500600700800
50 100 150 200 250 Time Step
No. ofCPUs
No. of QM Atoms
270
No. of CPUs/Atoms
Succeeded in long-run by Succeeded in long-run by intentional/unintentional resource intentional/unintentional resource migrationmigration
Intentional migrationMigration triggered by faults
Summary of the experimental results
We could verify that our strategy for long run is praWe could verify that our strategy for long run is practical approachctical approach
Continue the simulation by migrating one cluster to the other one based on reservation
We could verify the programming using GridRPC anWe could verify the programming using GridRPC and MPI could implement real Grid-enabled applicatiod MPI could implement real Grid-enabled applicationn
Dynamic resource allocation / migrationRecover from faultsManage hundreds of CPUs on distributed sites
Status and Future Plans
Ninf-G Version 5 will be coming!Ninf-G Version 5 will be coming!What are differences with Ninf-G4?What are differences with Ninf-G4?
Lower prerequisites for installationNinf-G4 needs Globus Library since it uses Globus IO for client/server communications.Ninf-G5 can be installed without Globus. i.e.,
Ninf-G5 can be installed according to the underlying softwarNinf-G5 can be installed according to the underlying software environmentse environments
Three major components (remote process invocation, information retrieval, and client/server communication) can be pluggable. e.g. without Globus, without TCPWork efficiently from a single supercomputer to Grid
Other new features will be supportedOther new features will be supportedConnection less (client <-> server)Client-side check pointing
Ninf-G 5.0.0alpha will be available in this March.Ninf-G 5.0.0alpha will be available in this March.