- Title
- Why decide: is a user's estimation of job completion time useful in grid resource allocation?
- Creator
- Lynar, Timothy M.; Herbert, Ric D.; Simon; Chivers, William J.
- Relation
- 18th World IMACS Congress and MODSIM09 International Congress on Modelling and Simulation. The 18th World IMACS Congress and MODSIM09 International Congress on Modelling and Simulation: Proceedings (Cairns, Qld 13-17 July, 2009) p. 1031-1037
- Relation
- http://www.mssanz.org.au/modsim09/authorsL-M.htm#l
- Publisher
- Modelling and Simulation Society of Australia and New Zealand and International Association for Mathematics and Computers in Simulation
- Resource Type
- conference paper
- Date
- 2009
- Description
- There are many resource allocation mechanisms for grid computing. One trait that many allocation mechanisms have, is that they require the user to estimate how long their task will take to execute on a system. The result of the user’s input can have a considerable impact on scheduling, and can affect the grids ability to meet quality of service requirements for other jobs. Incorrect estimates by users could result in other jobs being turned away that should not have, or jobs being accepted, that should not have been. The user’s estimation could conceivably be accurate if the user is running the same jobs repeatedly on the same hardware. However this is rarely the case in a grid environment. This estimation is often an estimation that is made with limited or no knowledge of how long a task will take on the underlying hardware and can be considered to be a guess. In this paper we have tested the accuracy of user estimation of task execution, through simulating the grid environment. The simulation includes a simple agent-based batch auction to distribute jobs to resources. The simulated environment contains a number of resources. Tasks are submitted to the environment periodically by agents who estimate the length of time their job will take. Each agent has only one job but submits that job several times throughout the simulation. Each job requires a random amount of processing and each resource can process a random amount per time-step. There are three groups, each group with equal quantities of agents. Each group uses one of three strategies to estimate the length of time their task will take to execute on the grid. One strategy uses the agent’s limited memory to estimate the length of execution based on the time previous tasks took for that agent to execute over the grid; a second strategy utilises a history from the assigned resource as to how long previous jobs have taken to execute on the resource; and the third strategy is a zero intelligence strategy, it is a random guess. The results show whether there is any significant difference between the accuracy of an estimation made using historical data from a user, using historical data from the resource, or a random guess. The results have implications for the design of grid resource allocation mechanisms, and for how users interact with current resource allocation mechanisms. It also raised the question, when should a user’s input be required and on what will it be based?
- Subject
- resource allocation; grid computing; allocation mechanisms; task estimation
- Identifier
- http://hdl.handle.net/1959.13/919471
- Identifier
- uon:8880
- Identifier
- ISBN:9780975840078
- Language
- eng
- Reviewed
- Hits: 1555
- Visitors: 1783
- Downloads: 3
Thumbnail | File | Description | Size | Format |
---|