Computer Sciences Dept.

Joe Meehean

Former Graduate Student

Condor Cold Start: Dynamic Installation of the Condor Schedd (2005)


[PowerPoint]
This project developed dynamic deployment tools to allow new Condor pools to be installed and connected on-the-fly. By leveraging the lowest common denominator of grid middle-ware systems, simple program execution, and Condor's ingrained fault-tolerance these deployment tools are able to bind together heterogeneous grid resources with different management policies into a single dynamic Condor pool. The beach-head for each compute site is the dynamically deployed Condor scheduler which manages job submission and fault-tolerant scheduling. The computational units themselves are managed by the Condor resource management services that glide-in to create a new pool of Condor resources. Distribution of jobs to these new Condor sites is handled by the Condor-C (Condor-to-Condor) mechanism for transferring the responsibility for executing an application from one site to the next. Automatic and intelligent distribution of job scheduling across computation sites is a feature in active development, currently managed by a component that runs alongside the Condor scheduler. It is the portability

A Case for Dynamic File Attributes (CS736 Spring '04)


[PDF]
A file attribute is a user defined key-value pair associated with a file. We explore the idea of extending a file system to contain dynamic per-file attributes. This paper has two goals: first, it describes our implementation of an attribute file system overlay, in which the attribute functionality is provided as a user-level library and the system calls are modified to provide a seamless environment to users and applications. We compare the performance of our implementation under various scenarios, and we modify several applications to demonstrate the implications and benefits of customizable file attributes. Second, we explore the use of file attributes as a mechanism for approximating full file-content search. Our approach is to define a file-type-independent attribute format in which searchable data is stored, allowing for simple, type-agnostic indexing and search tools to be used. Preliminary performance and functionality tests indicate that this technique shows promise: searchable indexes may be built and searched with small disk-space and cpu-time overhead.

 
Computer Sciences | UW Home