reef-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Andrew Chung (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (REEF-1357) Allow different caching levels for caching in IInputPartition
Date Mon, 13 Jun 2016 21:25:03 GMT

    [ https://issues.apache.org/jira/browse/REEF-1357?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15328308#comment-15328308
] 

Andrew Chung commented on REEF-1357:
------------------------------------

Thanks for the input [~shravanmn]! Please correct me if I am wrong - are you suggesting that
we should only implement the caching logic for each {{IInputPartition}} as needed? The implementation
in this JIRA item abstracts out the caching portion of {{IInputPartition}} and does not necessarily
have to be used in tandem with {{IInputPartition}} itself. The levels of caching are entirely
defined by the user as integers, and so is the ability to move between different storage levels
(by registering different {{DataMover}} s. This allows the user to plug in different layers
of storage without modifying their implementation of {{IInputPartition}} - they would only
need to implement new {{DataMover}} s. Would you say that this abstraction is too complicated?

> Allow different caching levels for caching in IInputPartition
> -------------------------------------------------------------
>
>                 Key: REEF-1357
>                 URL: https://issues.apache.org/jira/browse/REEF-1357
>             Project: REEF
>          Issue Type: Improvement
>          Components: IMRU, REEF.NET
>            Reporter: Andrew Chung
>            Assignee: Andrew Chung
>              Labels: FT
>
> Caching in {{IInputPartition}} is currently entirely dependent on the implementation
of the {{IInputPartition}}. We should modify/expand {{IInputPartition}} such that the data
loading component is fully separate from the caching component. An idea is to instead provide
only a single concrete {{IInputPartition}} implementation (while deprecating the old implementations)
and use the composition model, complete with an {{IDataLoader}} which fetches data back in
a {{Stream}}, and an {{IDataCache}}, which allows caching at granular levels.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message