For a master only nodes on a small cluster (I mean small number of nodes/indices/mappings), you can probably set HEAP to 1gb. (512 Mb should work as well probably)
32GB for ES data
32GB for the OS for file caching?
And then there another 64GB to play with I supose 4GB for master node is ok.
But is it ok to deploy the master node on same box as data node on separate
JVM process since I have the horse power?
On Tuesday, 3 February 2015 10:48:10 UTC-5, David Pilato wrote:
For a master only nodes on a small cluster (I mean small number of
nodes/indices/mappings), you can probably set HEAP to 1gb. (512 Mb should
work as well probably)
And right i don't envision more than 10-12 node cluster
On Tuesday, 3 February 2015 16:45:01 UTC-5, John Smith wrote:
Eh well I have 128GB
32GB for ES data
32GB for the OS for file caching?
And then there another 64GB to play with I supose 4GB for master node is
ok.
But is it ok to deploy the master node on same box as data node on
separate JVM process since I have the horse power?
On Tuesday, 3 February 2015 10:48:10 UTC-5, David Pilato wrote:
For a master only nodes on a small cluster (I mean small number of
nodes/indices/mappings), you can probably set HEAP to 1gb. (512 Mb should
work as well probably)
It is fine to give 2-4 GB (though 4 is on higher side) to your masters as
they are not doing anything really heavy, all they do is exchange cluster
state/routing table etc.
Spinning off the masters on the same host (box) as data node can
potentially cause some issues (e.g. High CPU resulting in overall slowness
and master becoming unresponsive or host restart resulting in master as
well as data node loss). We have exactly same setup (infact we have even
beefy machines and during our topology design we kept masters on separate
hardware which can be commodity hardware)
If you can't have more hardware, then spreading master across multiple
hosts will mitigate the split brain to a great extent.
On Tuesday, 3 February 2015 07:09:44 UTC-8, John Smith wrote:
Hi I read bunch of things on master nodes etc...
So I have some decent machines 4x 32core 128GB RAM and 6x960GB SSDS (RAID
5 so 4.3 TB)
So all 4 machines will be data nodes configured to 32GB.
I was also thinking of running the master nodes on same machines on
separate JVM using N/2+1. But how much RAM do master nodes need to be etc?
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.