Computer Sciences Dept.

Joe Meehean

Former Graduate Student

Towards Reliable Storage Systems
Joseph T. Meehean
Doctoral Dissertation, August 2011
Available as: Abstract, PDF,


A Service Migration Case Study: Migrating the Condor Schedd


[PDF][BibTeX Source for Citation]
Service migration has become an important topic due the rising interest in both service- based architectures and mobile computing. We have identified two core problems asso- ciated with migrating a service: packaging the service binaries and data in a fashion that allows it to be restarted at a remote site and locating a service after it has migrated. Many implementations of service migration assume homogeneous host architectures as well as uniform file access. Additionally, some implementations require that migration occur in kernel-space. We require that a service capture its own state using configuration files and operation logs. This state is then marshalled to be machine architecture independent as well as independent of any file system or mount point. We call this technique service-defined logical checkpointing, it occurs entirely in user-space and significantly eases the migration of a service. We mobilized the Condor High Throughput System's distributed scheduling service (schedd) to illustrate the use of service-defined logical checkpointing to migrate a service. Further, we created a specialized Condor command and associated Condor job that can be used to migrate a schedd to a specific host or to a host matching an arbitrary set of requirements, including CPU load

Logical Image Migration Based on Overlays (Technical Report)


[PDF] [BibTeX Source for Citation] Virtual machine technology is becoming an increas- ingly popular vehicle for enabling process and service migration. Many migration frameworks rely on a distributed file system to avoid worrying about the burden of migrating a large virtual disk. While this assumption is valid in some contexts, there are other environments where it is not feasible. In these cases, disk migration becomes the primary bottleneck for achieving efficient migration. We discuss the design, implementation, and evaluation of LIMBO, a sys- tem for efficient migration of VM local file systems. We've found that overlay-based migration using a copy-on-write block device can significantly improve migration costs while imposing a small amount of overhead on file system operations.

 
Computer Sciences | UW Home