Low performance in cluster


(oleg_go) #1

Hi,
I made a program which use ElasticSearch for searching members. There are a lot of members in the list, more then sixty million. It means we need scalability, so I have to research how ElasticSearch would run in a cluster. There are six million members in a test environment and two equal hosts under Windows7. There are ten threads in a .NET program for load test they send unique search query to ElacticSearch. And I get unexpected results when I use the clustered ElasticSearch. If I use one node and ElacticSearch runs on the one host I get some results. CPU utilization on the host is 95 percent, there is some average time to response and some number requests per second. After another host is added in the cluster there are worse results in test. CPU utilization on the both hosts in the cluster has got twenty percent, but average time to response and number requests per second has become two times worse. I don't understand how it could be. Are there any ideas?


(oleg_go) #2

In addition, when I added third node without data "node.data: false" on the host where client run and connected to it, performance in the cluster became better. CPU utilization in the hosts became about seventy percent, and time to response and number requests per second had become better than with one node (without cluster). Would you answer to my question, please.


(Shay Banon) #3

Seems like the mailing list got only this message, can you report your
original message?

On Mon, Nov 14, 2011 at 1:26 PM, oleg_go ogoryachev@hill30.com wrote:

In addition, when I added third node without data "node.data: false" on the
host where client run and connected to it, performance in the cluster
became
better. CPU utilization in the hosts became about seventy percent, and time
to response and number requests per second had become better than with one
node (without cluster).

--
View this message in context:
http://elasticsearch-users.115913.n3.nabble.com/Low-performance-in-cluster-tp3499497p3506551.html
Sent from the ElasticSearch Users mailing list archive at Nabble.com.


(oleg_go) #4

Here it is...

Hi,
I made a program which use ElasticSearch for searching members. There are a lot of members in the list, more then sixty million. It means we need scalability, so I have to research how ElasticSearch would run in a cluster. There are six million members in a test environment and two equal hosts under Windows7. There are ten threads in a .NET program for load test they send unique search query to ElacticSearch. And I get unexpected results when I use the clustered ElasticSearch. If I use one node and ElacticSearch runs on the one host I get some results. CPU utilization on the host is 95 percent, there is some average time to response and some number requests per second. After another host is added in the cluster there are worse results in test. CPU utilization on the both hosts in the cluster has got twenty percent, but average time to response and number requests per second has become two times worse. I don't understand how it could be. Are there any ideas?


(Shay Banon) #5

First, can you try with 0.18.3. Also, are those search requests or indexing
request? If its indexing requests, adding another node (with a 1 replica
index) will make indexing slower, because they need to be replicated to
another node.

Its strange what you saw with adding a client node on the same box as the
"client". Maybe its simply that ES has better networking lib, so going
"localhost" from .NET to ES will be fast, and then ES will do the rest...

On Tue, Nov 15, 2011 at 8:08 PM, oleg_go ogoryachev@hill30.com wrote:

Here it is...

Hi,
I made a program which use ElasticSearch for searching members. There are a
lot of members in the list, more then sixty million. It means we need
scalability, so I have to research how ElasticSearch would run in a
cluster.
There are six million members in a test environment and two equal hosts
under Windows7. There are ten threads in a .NET program for load test they
send unique search query to ElacticSearch. And I get unexpected results
when
I use the clustered ElasticSearch. If I use one node and ElacticSearch runs
on the one host I get some results. CPU utilization on the host is 95
percent, there is some average time to response and some number requests
per
second. After another host is added in the cluster there are worse results
in test. CPU utilization on the both hosts in the cluster has got twenty
percent, but average time to response and number requests per second has
become two times worse. I don't understand how it could be. Are there any
ideas?

--
View this message in context:
http://elasticsearch-users.115913.n3.nabble.com/Low-performance-in-cluster-tp3499497p3510474.html
Sent from the ElasticSearch Users mailing list archive at Nabble.com.


(system) #6