giraph-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Francesco Sclano <>
Subject In a real cluster (not pseudo-distributed mode) composed by N datanodes eachone having X vCPU, is more appropriate to set giraph.numComputeThreads with N or with X*N ?
Date Wed, 12 Dec 2018 15:31:08 GMT
I'm using a cluster composed of N datanodes and I'd like to properly set
following parameters:

-ca giraph.numComputeThreads=
-ca giraph.numInputThreads=
-ca giraph.numOutputThreads=

The cluster is dedicated to my giraph program (no others job run in this
I'd like to know if is more appropriate to set above parameters with N *
with X*N ? (where X is the number of vCPU in every datanode).

p.s. As Claudio Martella says in a previous post, maybe is better to use
X-1 instead of X because giraph needs 1 thread for internal work. Anyway I
didn't understand if this previous post is related to a pseudo-distributed
mode or a distributed mode. Hence i post this question :-P


Francesco Sclano

View raw message