A simple virtual organisation model and practical implementation - simulation visualised
The VO managed job queueing service prototype was
tested in a simulation together with the previously
mentioned resource level job consumer. The simulation
included 3 VOs each with 10 users of varying VO
priority. The VOs maintained one queueing service
each. All VOs have access to 10 simulated resources,
each consisting of a clusters of 10 to 50 nodes or job
execution slots. Each resource allocated a different
local priority range for each VO queue. The ranges
were of at least 10 and randomly allocated between
1 and 100. Each user periodically submitted between
10 and 50 jobs at once but only when they had no
jobs in the queue. Each job took between 1 and 10
minutes to complete on any node.
This particular scenario led to an average stable
state of 30 jobs for each of 30 users. These were shared
across 10 hosts with an average of 30 nodes. The
average job load for this situation is 3 times more
than the number of available nodes, an over utilised
resource scenario. Resource usage reached saturation
within 4 minutes of the simulation commencing. The
state of the queue was frozen after 15 minutes for
evaluation.
tested in a simulation together with the previously
mentioned resource level job consumer. The simulation
included 3 VOs each with 10 users of varying VO
priority. The VOs maintained one queueing service
each. All VOs have access to 10 simulated resources,
each consisting of a clusters of 10 to 50 nodes or job
execution slots. Each resource allocated a different
local priority range for each VO queue. The ranges
were of at least 10 and randomly allocated between
1 and 100. Each user periodically submitted between
10 and 50 jobs at once but only when they had no
jobs in the queue. Each job took between 1 and 10
minutes to complete on any node.
This particular scenario led to an average stable
state of 30 jobs for each of 30 users. These were shared
across 10 hosts with an average of 30 nodes. The
average job load for this situation is 3 times more
than the number of available nodes, an over utilised
resource scenario. Resource usage reached saturation
within 4 minutes of the simulation commencing. The
state of the queue was frozen after 15 minutes for
evaluation.
