Research On Task Migration Method Based On Checkpoint

Task migration is a widely used technology in cluster systems.This technology can improve the reliability and fault tolerance of the system,achieve dynamic load balancing of the system and improve the utilization of system resources.Task migration is divided into independent process migration and dependent process migration according to the objects being migrated.Checkpoint technology is one of the main ways to achieve process migration.At present,the process of migration based on checkpoint has the problems that the communication after dependent process migrationt cannot be handled,the time and space overhead of process migration are large,the process migration operation across heterogeneous platforms is complicated and the source code is modified a lot.In order to solve the above problems,the main work and contributions of this thesis are including as follow:(1)Based on the analysis of the representative checkpoint tools Condor,BLCR,WinNtCkpt,and NTckpt on the Linux and Windows platforms,the advantages and disadvantages of the collaborative checkpoint algorithm and non-cooperative checkpoint algorithm are summarized and the typical collaborative checkpoint algorithms "Sync-and-stop" and "Chandy-Lamport" are compared to provide a globally consistent checkpoint operation for migration of dependent processes.(2)In view of the problem of restoring communication in migration of dependent processes,the migration methods based on checkpoint tools of different platforms are proposed,and the feasibility of the methods are verified through experiments.The checkpoint tool CKPT is further proposed to achieve the migration across heterogeneous platforms,and verified the effectiveness of the tool through experiments.(3)For different migration scenarios,task migration systems based on checkpoint are designed and implemented,and the migration methods proposed in this thesis are evaluated experimentally.Experimental results show that the proposed methods can achieve task migration with lower time and space overhead.