[
https://issues.apache.org/jira/browse/IGNITE-27279?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Alexey Scherbakov updated IGNITE-27279:
---------------------------------------
Description:
Currently AI3 uses hard-coded default zone partition count [1]
Instead, partition count should be calculated dynamically, using the formula:
{*}nodes_count * max(cores_count, 8) * scale_coeff{*} / replica_cnt, where
* nodes_count is the count of nodes in initial topology
* cores_count is the number of cores per node
* replica_cnt is the number of replicas
* scale_coeff defines cluster upscaling, with default value of 3. Can also
test a value of 4 - this should give better assignments, but performance
overhead per partition must be taken into consideration.
[1]
org.apache.ignite.internal.catalog.commands.CatalogUtils#DEFAULT_PARTITION_COUNT
+ info if the distribution occurred to be inefficient.
was:
Currently AI3 uses hard-coded default zone partition count [1]
Instead, partition count should be calculated dynamically, using the formula:
{*}nodes_count * max(cores_count, 8) * scale_coeff{*}, where
* nodes_count is the count of nodes in initial topology
* cores_count is the number of cores per node
* scale_coeff defines cluster upscaling, with default value of 2. Can also
test values of 3 or 4 - this should give better assignments, but performance
overhead per partition must be taken into consideration.
[1]
org.apache.ignite.internal.catalog.commands.CatalogUtils#DEFAULT_PARTITION_COUNT
+ info if the distribution occurred to be inefficient.
> Change default partition count
> ------------------------------
>
> Key: IGNITE-27279
> URL: https://issues.apache.org/jira/browse/IGNITE-27279
> Project: Ignite
> Issue Type: Improvement
> Reporter: Alexey Scherbakov
> Assignee: Mikhail Efremov
> Priority: Major
> Labels: ignite-3
> Fix For: 3.2
>
>
> Currently AI3 uses hard-coded default zone partition count [1]
> Instead, partition count should be calculated dynamically, using the formula:
> {*}nodes_count * max(cores_count, 8) * scale_coeff{*} / replica_cnt, where
> * nodes_count is the count of nodes in initial topology
> * cores_count is the number of cores per node
> * replica_cnt is the number of replicas
> * scale_coeff defines cluster upscaling, with default value of 3. Can also
> test a value of 4 - this should give better assignments, but performance
> overhead per partition must be taken into consideration.
> [1]
> org.apache.ignite.internal.catalog.commands.CatalogUtils#DEFAULT_PARTITION_COUNT
> + info if the distribution occurred to be inefficient.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
