Renkai,
The execution models between storm and akka are rather different. Storm has
dedicated threads for each executor, although multiple tasks (bolt/spout
instances) can share the same executor thread. We rely on the OS to do the
right thing in scheduling threads to run. Akka is using an actor model. The
dispatcher essentially provides a thread pool for all of the actors to share.
Instead of allowing the OS to do all of the context switching they use the
dispatcher to decide in many cases when one actor should stop executing and
another should start. Currently storm tends to run with latency in mind over
throughput, but we are working on adding in tuning knobs to allow for more
throughput at the expense of some latency. However the ultimate goal is to get
to the point where you tell storm the constraints you have in terms of latency,
throughput, and cost and let storm work to optimize the topology to meet those
desires. This is still a ways out, but to me feels much more user friendly
then trying to figure out what the batch size should be for a given actor
manually.
- Bobby
On Sunday, October 18, 2015 10:03 AM, Renkai <[email protected]> wrote:
Hi all:
I am a storm user for one year and now I am learning Akka. I found something
interesting in Akka that it could config its dispatcher(see
http://doc.akka.io/docs/akka/2.4.0/scala/dispatchers.html). I think it is
useful for tuning performance and storm seems missing something like that.
