[
https://issues.apache.org/jira/browse/DRILL-5266?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15868657#comment-15868657
]
Paul Rogers commented on DRILL-5266:
------------------------------------
The result of the "bits vs. bytes" fix improves density to 28%:
{code}
Actual batch schema & sizes {
T1¦¦cs_sold_date_sk(std col. size: 4, actual col. size: 4, total size:
196608, vector size: 131072, data size: 36156, row capacity: 32768, density: 28)
...
T1¦¦c_customer_id(std col. size: 54, actual col. size: 16, total size:
2392064, vector size: 2359296, data size: 144624, row capacity: 32768, density:
7)
...
T1¦¦c_salutation(std col. size: 54, actual col. size: 4, total size: 2392064,
vector size: 2359296, data size: 28544, row capacity: 32768, density: 2)
T1¦¦c_first_name(std col. size: 54, actual col. size: 6, total size: 2392064,
vector size: 2359296, data size: 50916, row capacity: 32768, density: 3)
T1¦¦c_last_name(std col. size: 54, actual col. size: 6, total size: 2392064,
vector size: 2359296, data size: 53546, row capacity: 32768, density: 3)
T1¦¦c_preferred_cust_flag(std col. size: 54, actual col. size: 1, total size:
2392064, vector size: 2359296, data size: 8780, row capacity: 32768, density: 1)
...
T1¦¦c_login(std col. size: 54, actual col. size: 0, total size: 2392064,
vector size: 2359296, data size: 0, row capacity: 32768, density: 0)
T1¦¦c_email_address(std col. size: 54, actual col. size: 27, total size:
2392064, vector size: 2359296, data size: 238285, row capacity: 32768, density:
11)
...
c_email_address(std col. size: 54, actual col. size: 27, total size: 344064,
vector size: 327680, data size: 238285, row capacity: 16383, density: 73)
Records: 9039, Total size: 32325632, Row width:3577, Density:26}
{code}
The most dense vector is now 73%, but most are 28%.
Performance has improved from 37 secs to 27 secs:
{code}
Results: 1,434,519 records, 30 batches, 26,612 ms
{code}
> Parquet Reader produces "low density" record batches
> ----------------------------------------------------
>
> Key: DRILL-5266
> URL: https://issues.apache.org/jira/browse/DRILL-5266
> Project: Apache Drill
> Issue Type: Bug
> Components: Storage - Parquet
> Affects Versions: 1.10
> Reporter: Paul Rogers
>
> Testing with the managed sort revealed that, for at least one file, Parquet
> produces "low-density" batches: batches in which only 5% of each value vector
> contains actual data, with the rest being unused space. When fed into the
> sort, we end up buffering 95% of wasted space, using only 5% of available
> memory to hold actual query data. The result is poor performance of the sort
> as it must spill far more frequently than expected.
> The managed sort analyzes incoming batches to prepare good memory use
> estimates. The following the the output from the Parquet file in question:
> {code}
> Actual batch schema & sizes {
> T1¦¦cs_sold_date_sk(std col. size: 4, actual col. size: 4, total size:
> 196608, vector size: 131072, data size: 4516, row capacity: 32768, density: 4)
> T1¦¦cs_sold_time_sk(std col. size: 4, actual col. size: 4, total size:
> 196608, vector size: 131072, data size: 4516, row capacity: 32768, density: 4)
> T1¦¦cs_ship_date_sk(std col. size: 4, actual col. size: 4, total size:
> 196608, vector size: 131072, data size: 4516, row capacity: 32768, density: 4)
> ...
> c_email_address(std col. size: 54, actual col. size: 27, total size: 53248,
> vector size: 49152, data size: 30327, row capacity: 4095, density: 62)
> Records: 1129, Total size: 32006144, Row width:28350, Density:5}
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
