I'm benchmarking results for an ES cluster, using both the REST api and
native Java client. We're getting very different response times, between
each of these (the REST api is doing approximately 50% better in the 95th
percentile). I was wondering what the cause of this is, since the search
requests look identical to me (unless I missed something).
Here are the queries:
Our setup: 20 node cluster, 20 shards/2 replicas, and benchmarking from a
separate machine with JMeter (an HTTP sampler for each of the 20 nodes, and
a separate test plan with a Java sampler initializing the ES java client,
and sending queries).
I'm benchmarking results for an ES cluster, using both the REST api and
native Java client. We're getting very different response times, between
each of these (the REST api is doing approximately 50% better in the 95th
percentile). I was wondering what the cause of this is, since the search
requests look identical to me (unless I missed something).
Our setup: 20 node cluster, 20 shards/2 replicas, and benchmarking from a
separate machine with JMeter (an HTTP sampler for each of the 20 nodes, and
a separate test plan with a Java sampler initializing the ES java client,
and sending queries).
I tried that, and it seems to make no difference. All I could gather are
these graphs (attached) from the bigdesk plugin that are showing
performance for the query and fetch phase (the first is during ES java
client use, the second is using the REST calls) . It's really weird that
the FETCH phase is taking any time at all, if I set _source = false.
On Thursday, December 18, 2014 5:33:05 PM UTC-5, Jörg Prante wrote:
On Thu, Dec 18, 2014 at 7:48 PM, Marie Jacob <jacob...@gmail.com
<javascript:>> wrote:
I'm benchmarking results for an ES cluster, using both the REST api and
native Java client. We're getting very different response times, between
each of these (the REST api is doing approximately 50% better in the 95th
percentile). I was wondering what the cause of this is, since the search
requests look identical to me (unless I missed something).
Our setup: 20 node cluster, 20 shards/2 replicas, and benchmarking from a
separate machine with JMeter (an HTTP sampler for each of the 20 nodes, and
a separate test plan with a Java sampler initializing the ES java client,
and sending queries).
I tried that, and it seems to make no difference. All I could gather are
these graphs (attached) from the bigdesk plugin that are showing
performance for the query and fetch phase (the first is during ES java
client use, the second is using the REST calls) . It's really weird that
the FETCH phase is taking any time at all, if I set _source = false.
On Thursday, December 18, 2014 5:33:05 PM UTC-5, Jörg Prante wrote:
I'm benchmarking results for an ES cluster, using both the REST api and
native Java client. We're getting very different response times, between
each of these (the REST api is doing approximately 50% better in the 95th
percentile). I was wondering what the cause of this is, since the search
requests look identical to me (unless I missed something).
Our setup: 20 node cluster, 20 shards/2 replicas, and benchmarking from
a separate machine with JMeter (an HTTP sampler for each of the 20 nodes,
and a separate test plan with a Java sampler initializing the ES java
client, and sending queries).
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.