based on observations and tests which show that exceptionally long task
pools are very slow to execute.
We now chunk tasks such that each there are ~10 tasks per available
processor in each chunk. This is an attempt to balance efficiency
against the limitation identified here.
based on observations and tests which show that exceptionally long task pools are very slow to execute.
We now chunk tasks such that each there are ~10 tasks per available processor in each chunk. This is an attempt to balance efficiency against the limitation identified here.