cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jonathan Ellis (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-7065) Add some extra metadata in leveled manifest to be able to reduce the amount of sstables searched on read path
Date Tue, 22 Apr 2014 16:26:16 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-7065?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13976999#comment-13976999
] 

Jonathan Ellis commented on CASSANDRA-7065:
-------------------------------------------

I think we already have the info we need in the SSTR objects, but IntervalTree is level-agnostic
so doesn't take advantage of it.  (We could in theory have the best of both worlds.)

> Add some extra metadata in leveled manifest to be able to reduce the amount of sstables
searched on read path
> -------------------------------------------------------------------------------------------------------------
>
>                 Key: CASSANDRA-7065
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-7065
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: Marcus Eriksson
>
> Based on this;
> http://rocksdb.org/blog/431/indexing-sst-files-for-better-lookup-performance/
> By keeping pointers from the sstables in lower to higher levels we could reduce the number
of candidates in higher levels, ie, instead of searching all 1000 L3 sstables, we use the
information from the L2 search to include less L3 sstables.
> First we need to figure out if this can beat our IntervalTree approach (and if the win
is worth it).



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Mime
View raw message