[
https://issues.apache.org/jira/browse/PHOENIX-1035?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14238677#comment-14238677
]
Mujtaba Chohan commented on PHOENIX-1035:
-----------------------------------------
[~jamestaylor] Performance issue is still there with local index. Performance
is ~50X slower for count(*) query compared to mutable index and ~25X slower a
min or two after data load.
Here is the schema I used for testing with 3M rows upserted.
{code}
CREATE TABLE LOCAL_INDEXED_TABLE_WIDE (HOST CHAR(2) NOT NULL,DOMAIN CHAR(200)
NOT NULL,
FEATURE VARCHAR NOT NULL,DATE DATE NOT NULL,USAGE.CORE BIGINT,USAGE.DB
BIGINT,STATS.ACTIVE_VISITOR INTEGER, STATS.NONINDEXED
INTEGER CONSTRAINT PK PRIMARY KEY (HOST, DOMAIN, FEATURE, DATE))
MAX_FILESIZE=30485760;
CREATE LOCAL INDEX lidx1 ON LOCAL_INDEXED_TABLE_WIDE (CORE);
CREATE LOCAL INDEX lidx2 ON LOCAL_INDEXED_TABLE_WIDE (DB);
CREATE LOCAL INDEX lidx3 ON LOCAL_INDEXED_TABLE_WIDE (DB,ACTIVE_VISITOR);
CREATE LOCAL INDEX lidx4 ON LOCAL_INDEXED_TABLE_WIDE (CORE,DB,ACTIVE_VISITOR);
{code}
Performance numbers with Local Index:
{code}
select count(*) from LOCAL_INDEXED_TABLE_WIDE - query time:[33645ms]
EXPLAIN: CLIENT 102-CHUNK PARALLEL 32-WAY RANGE SCAN OVER
_LOCAL_IDX_LOCAL_INDEXED_TABLE_WIDE [-32768]
SERVER AGGREGATE INTO SINGLE ROW
{code}
Performance numbers with Mutable Index:
{code}
select count(*) from MUTABLE_INDEXED_TABLE_WIDE - query time:[582ms]
EXPLAIN: CLIENT 107-CHUNK PARALLEL 1-WAY FULL SCAN OVER MIDX1
SERVER AGGREGATE INTO SINGLE ROW
{code}
2. Another issue is that local index table did not end up on the same region
server as the base table.
HBase v0.98.7-hadoop2
Phoenix v4.2.2 (commit hash: 5c6fc2f02d01805255fff335abb675ece07d07d0)
> Performance test local indexing against global indexing
> -------------------------------------------------------
>
> Key: PHOENIX-1035
> URL: https://issues.apache.org/jira/browse/PHOENIX-1035
> Project: Phoenix
> Issue Type: Sub-task
> Reporter: James Taylor
> Assignee: Mujtaba Chohan
>
> Local indexing is close to being in. We should perf test it both for reads
> and writes against global secondary indexing.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
