[
https://issues.apache.org/jira/browse/IGNITE-18977?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Vladislav Pyatkov updated IGNITE-18977:
---------------------------------------
Description:
*Motivation*
In a distributed system, all nodes have their own time view. A task to take
current time on a particular node have no guarantee that the timestamp is
achieved in all members in the topology. But our distributed algorithms
required to time synchronization (for example [transaction
protocol|https://cwiki.apache.org/confluence/display/IGNITE/IEP-91%3A+Transaction+protocol#IEP91:Transactionprotocol-Clockrequirements])
until a specific accuracy.
The time interval on which all cluster clock can different is called a clock
skew. Cluster should support whatever any cluster node has the clock skew does
not bigger that it is defined for the topology.
was:
*Motivation*
In a distributed system, all nodes have their own time view. A task to take
current time on a particular node have no guarantee that the timestamp is
achieved in all members in the topology. But our distributed algorithms
required to time synchronization until a specific accuracy.
The time interval on which all cluster clock can different is called a clock
skew. Cluster should support whatever any cluster node has the clock skew does
not bigger that it is defined for the topology.
> Clock skew managment
> --------------------
>
> Key: IGNITE-18977
> URL: https://issues.apache.org/jira/browse/IGNITE-18977
> Project: Ignite
> Issue Type: Epic
> Reporter: Vladislav Pyatkov
> Priority: Major
> Labels: ignite-3
>
> *Motivation*
> In a distributed system, all nodes have their own time view. A task to take
> current time on a particular node have no guarantee that the timestamp is
> achieved in all members in the topology. But our distributed algorithms
> required to time synchronization (for example [transaction
> protocol|https://cwiki.apache.org/confluence/display/IGNITE/IEP-91%3A+Transaction+protocol#IEP91:Transactionprotocol-Clockrequirements])
> until a specific accuracy.
> The time interval on which all cluster clock can different is called a clock
> skew. Cluster should support whatever any cluster node has the clock skew
> does not bigger that it is defined for the topology.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
