You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Nested parallelism is probably desirable, ie, a worker should be allowed to invoke Par.invoke(), suspending until that subcomputation is done. Such a facility would allow computations to be subdivided recursively.
(Broadcast and eval are in a sense less obvious because they modify global state in the workers. Invoke can do that too, but doesn't have to.)
Recursive invoke() can probably be implemented by having a notion of nested work sets, one work set at each nesting level. A worker might grab a work item from the innermost set, until that set is exhausted, or might grab a work item from a set with some probability that is higher for inner sets and lower for outer sets. Presumably work has been done on this kind of problem before.
The text was updated successfully, but these errors were encountered:
Nested parallelism is probably desirable, ie, a worker should be allowed to invoke Par.invoke(), suspending until that subcomputation is done. Such a facility would allow computations to be subdivided recursively.
(Broadcast and eval are in a sense less obvious because they modify global state in the workers. Invoke can do that too, but doesn't have to.)
Recursive invoke() can probably be implemented by having a notion of nested work sets, one work set at each nesting level. A worker might grab a work item from the innermost set, until that set is exhausted, or might grab a work item from a set with some probability that is higher for inner sets and lower for outer sets. Presumably work has been done on this kind of problem before.
The text was updated successfully, but these errors were encountered: