[
https://issues.apache.org/jira/browse/IGNITE-17345?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Nikolay Izhikov updated IGNITE-17345:
-------------------------------------
Description:
The crucial point to understand ThinClient performance is to know - Partition
Awareness enabled or not.
For now, it's impossible to understand how many request goes to node that is
primary for key.
It seems useful metrics to analyze PA behavior - two counters to track amount
of requests for each server node
- one counter for keys current node is primary.
- another counter for keys which require extra network hop between server nodes
to serve the request.
For environment with optimal performance second counter should be close to zero.
was:
The crucial point to understand ThinClient performance is to know - Partition
Awareness
For now, it's impossible to understand how many request goes to node that is
primary for key.
It seems usefull metric for this - two counters to track amount of requests for
each server node
- one counter for keys current node is primary.
- another counter for keys which require extra network hop between server nodes
to serve the request.
For environment with optimal performance second counter should be close to zero.
> [IEP-35] Metric to track PA enabled request on ThinClient
> ---------------------------------------------------------
>
> Key: IGNITE-17345
> URL: https://issues.apache.org/jira/browse/IGNITE-17345
> Project: Ignite
> Issue Type: Improvement
> Reporter: Nikolay Izhikov
> Priority: Major
> Labels: IEP-35, ise
>
> The crucial point to understand ThinClient performance is to know - Partition
> Awareness enabled or not.
> For now, it's impossible to understand how many request goes to node that is
> primary for key.
> It seems useful metrics to analyze PA behavior - two counters to track amount
> of requests for each server node
> - one counter for keys current node is primary.
> - another counter for keys which require extra network hop between server
> nodes to serve the request.
> For environment with optimal performance second counter should be close to
> zero.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
