Re: Need Help regarding topology with numWorker>1

2016-02-13 Thread Rudraneel chakraborty
Thank you And i need this because i am developing a custom scheduler . And for this i need to have topologies performing well with two worker processes instead of one. And its becoming tough to saturate a single worker On Saturday, 13 February 2016, Matthias J. Sax wrote: > If your workload does

Re: Need Help regarding topology with numWorker>1

2016-02-13 Thread Matthias J. Sax
If your workload does not saturate a single machine, it will of course be more efficient to run within a single worker as you avoid intra-JVM communication. As long as CPU, memory, or network of a single machine is not utilized completely, you will not benefit from multiple workers from a performa

Re: Need Help regarding topology with numWorker>1

2016-02-12 Thread Rudraneel chakraborty
Thanks for the reply. So after some thinking , i figured if i shorten the size of the worker incoming queue , i can create the scenario where using more than one worker might result in better performance. Any thoughts on how to do that? The property topology, receiver.buffer.size should be the the

Re: Need Help regarding topology with numWorker>1

2016-02-12 Thread Rudraneel chakraborty
Thanks for the reply. So after some thinking , i figured if i shorten the size of the worker incoming queue , i can create the scenario where using more than one worker might result in better performance. Any thoughts on how to do that? The property topology, receiver.buffer.size should be the the

Re: Need Help regarding topology with numWorker>1

2016-02-12 Thread Rudraneel chakraborty
Thanks. Actually i have to create a scenario where 2 worker performs better than one worker. But in reality , topology with a single worker performs considerably better. I sending csv lines to kafka (5 partitions ) and reading them from a topology with kafka spout (parallelism hint 5) Any thought

Re: Need Help regarding topology with numWorker>1

2016-02-11 Thread Nathan Leung
Any situation where you require more CPU than 1 server can provide for you - there are tuning parameters (e.g. localOrShuffleGrouping) that you can use to reduce the amount of data sent over the network too. Any situation where you need to have tolerance in case of machine failure. On Thu, Feb

Re: Need Help regarding topology with numWorker>1

2016-02-11 Thread Rudraneel chakraborty
Topology param numWorkers i meant On Thursday, 11 February 2016, Rudraneel chakraborty < rudraneel.chakrabo...@gmail.com> wrote: > More specifically , i have seen a topology performs better if it is > assigned a single worker compared to more than one worker. > > I want a situation where a topolo

Re: Need Help regarding topology with numWorker>1

2016-02-11 Thread Rudraneel chakraborty
More specifically , i have seen a topology performs better if it is assigned a single worker compared to more than one worker. I want a situation where a topology performs better with more than one worker. And it doesnt matter if both workers are on same supervisor or different supervisor On Thu

Re: Need Help regarding topology with numWorker>1

2016-02-11 Thread Matthias J. Sax
I am not sure what you mean: - number of worker slots per supervisor or - topology parameter "number of workers" Can you clarify? -Matthias On 02/11/2016 05:14 AM, anshu shukla wrote: > Not like that.. But i have used workers equal to number of cores. Each > vm with 8 corea. > > On 11 Fe

Re: Need Help regarding topology with numWorker>1

2016-02-10 Thread anshu shukla
Not like that.. But i have used workers equal to number of cores. Each vm with 8 corea. On 11 Feb 2016 9:07 am, "Rudraneel chakraborty" < rudraneel.chakrabo...@gmail.com> wrote: > more than one worker on same node ? Did u use custom schedule r? because > by default, the workers would be spread t

Re: Need Help regarding topology with numWorker>1

2016-02-10 Thread Rudraneel chakraborty
more than one worker on same node ? Did u use custom schedule r? because by default, the workers would be spread throughout the cluster

Re: Need Help regarding topology with numWorker>1

2016-02-10 Thread anshu shukla
Since every worker process have only one thread for transfer of message from network to executor queue so more often it becomes a bottkeneck when input rate is high.. That forces us to have more than 1 worker on same node. I dont think other than this there is any logic dependent topo case whe

Need Help regarding topology with numWorker>1

2016-02-10 Thread Rudraneel chakraborty
Hello Good People, I desperately need an example of a topology which performs better with more than one worker process compared to a single worker. Could anyone help