Handle memouts more appropriately #27

New Issue

2017-11-27T20:54:01+01:00

patrick commented

2017-11-27 20:54:01 +01:00

In order to treat benchmark runs with memouts more justly, we should configure the corresponding configurations to respect the memory limits. In this way, there won’t be memouts, and the results might be more representative.

patrick added this to the Extended Paper for TPLP Journal milestone 2017-11-27 20:54:01 +01:00

patrick added the bug benchmarks task labels 2017-11-27 20:54:01 +01:00

patrick commented

2017-11-29 19:54:28 +01:00

I modified the configuration as follows:

The planners get a soft memory limit of 8192 MB. They should stick to this limit and not allocate more than that.
There is a tolerance margin of 1024 MB, meaning that the hard memory limit is 9216 MB. After that, the benchmark job is shut down, and the result is counted as a memout.

In this way, all the existing results are still valid.

I removed all results of affected instances (with memouts) and restarted the benchmark runner to redo them.

I modified the configuration as follows: - The planners get a soft memory limit of 8192 MB. They should stick to this limit and not allocate more than that. - There is a tolerance margin of 1024 MB, meaning that the hard memory limit is 9216 MB. After that, the benchmark job is shut down, and the result is counted as a memout. In this way, all the existing results are still valid. I removed all results of affected instances (with memouts) and restarted the benchmark runner to redo them.

patrick closed this issue

2017-11-29 19:54:28 +01:00

This repo is archived. You cannot comment on issues.

1 Participants

Due Date

-

Dependencies

No dependencies set.

Reference: patrick/plasp#27