Github user markhamstra commented on the pull request:
https://github.com/apache/spark/pull/27#issuecomment-36277558
I see two issues: 1) The deterministic nature of the current scheduler
places tasks on the same small set of machines while leaving others largely
unused; 2) There is no rebalancing of partitions across worker nodes when new
nodes are added to the cluster. Neither LRU nor randomization really addresses
the rebalancing issue, and LRU is only a little better than randomization in
addressing the unused workers issue, so I think the additional complexity of
LRU weighs against it -- at least until such time as we have evidence that
random isn't adequate.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
