[jira] [Resolved] (IGNITE-5075) Implement logical 'cache groups' sharing the same physical caches

["Semen Boikov (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/IGNITE-5075?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Semen Boikov resolved IGNITE-5075.
----------------------------------
    Resolution: Fixed

On public API only one new property 'groupName' was added for CacheConfiguration. Caches with the same groupName will share the same 'physical' cache partitions. Since caches in the same group shared the same partitions have should the same settings related to partitions assignment:
- affinity function
- node filter
- cache mode
- topology validation
- partition loss policy
- memory policy

In internal data structures we need distinguish same keys for different caches in the same groups. For this purpose in BPlusTree hash index we also store key's cacheId (int), this cacheId particiaptes in sorting and allows fast cache iteration. There are also on-heap maps in DhtLocalPartition, if cache group have multiple caches, then there will be separate maps for each cache, probably this can be optimzied.

> Implement logical 'cache groups' sharing the same physical caches
> -----------------------------------------------------------------
>
>                 Key: IGNITE-5075
>                 URL: https://issues.apache.org/jira/browse/IGNITE-5075
>             Project: Ignite
>          Issue Type: Task
>          Components: cache
>            Reporter: Semen Boikov
>            Assignee: Semen Boikov
>            Priority: Critical
>              Labels: important
>             Fix For: 2.1
>
>
> Currently started caches have pretty large memory overhead (e.g. to store affinity and partitions state information). If some application requires thousands caches with similar configuration then it would be useful to allow for 'logical' cache to reuse the same 'physical' cache. Let's introduce new CacheConfiguration property - 'groupName', caches with the same groupName will use the same 'physical' cache.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)