asterixdb-notifications mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF subversion and git services (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (ASTERIXDB-2186) Cache-friendly bloom filter
Date Fri, 15 Dec 2017 22:33:00 GMT

    [ https://issues.apache.org/jira/browse/ASTERIXDB-2186?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16293371#comment-16293371
] 

ASF subversion and git services commented on ASTERIXDB-2186:
------------------------------------------------------------

Commit ed503080786d6ab0694b68b0965f18edad9ceaa5 in asterixdb's branch refs/heads/master from
[~luochen01]
[ https://git-wip-us.apache.org/repos/asf?p=asterixdb.git;h=ed50308 ]

[ASTERIXDB-2186][STO] Cache-friendly Bloom Filter

- user model changes: no
- storage format changes: yes. Change the format of bloom filter
- interface changes: no

Details:
- Introduce blocked bloom filter to guarantee only 1 random memory
access is required during each membership check. It improves bloom
filter performance by 2x - 4x, depending on the ratio of positive
queries.
- For legacy bloom filters, we fall back to previous implementation
based on the stored version in the metadata.
- Add pinAllPages/unpinAllPages method to reduce pin/unpin overhead.

Change-Id: I0e8e0db9b60d5addfaf61ebb372a1bcb2d2d5957
Reviewed-on: https://asterix-gerrit.ics.uci.edu/2201
Sonar-Qube: Jenkins <jenkins@fulliautomatix.ics.uci.edu>
Integration-Tests: Jenkins <jenkins@fulliautomatix.ics.uci.edu>
Tested-by: Jenkins <jenkins@fulliautomatix.ics.uci.edu>
Contrib: Jenkins <jenkins@fulliautomatix.ics.uci.edu>
Reviewed-by: Ian Maxon <imaxon@apache.org>


> Cache-friendly bloom filter
> ---------------------------
>
>                 Key: ASTERIXDB-2186
>                 URL: https://issues.apache.org/jira/browse/ASTERIXDB-2186
>             Project: Apache AsterixDB
>          Issue Type: Improvement
>          Components: STO - Storage
>            Reporter: Chen Luo
>            Assignee: Chen Luo
>
> The current implementation of bloom filter is not cache friendly. A membership check
needs to perform multiple hash probes, each of which requires a separate random memory access
(i.e., cache miss).



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message