[
https://issues.apache.org/jira/browse/DRILL-6814?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16735852#comment-16735852
]
Denys Ordynskiy commented on DRILL-6814:
----------------------------------------
Comparison of the response time for querying tables on S3 and MapRFS storage.
Drill setups were on my local virtual machine "KVM" and EC2 instance.
*Response time for querying big parquet tables on S3 is not longer then on
MapRFS storage.*
Querying text files:
KVM S3 avg time - 369.22 seconds
KVM MapRFS avg time - 74.54 seconds
EC2 S3 avg time - 102.94 seconds
EC2 MapRFS avg time - 94.65 seconds
Querying parquet files:
KVM S3 avg time - 356.36 seconds
KVM MapRFS avg time - 79.54 seconds
EC2 S3 avg time - 114.47 seconds
EC2 MapRFS avg time - 105.46 seconds
| File format | EC2 MapRFS | EC2 S3 | KVM MapRFS | KVM S3 |
| CSV | 94.65 seconds | 102.94 seconds | 74.54 seconds | 369.22 seconds |
| PARQUET | 105.46 seconds | 114.47 seconds | 79.54 seconds | 356.36 seconds |
More details, profiles, in the attachment "S3investigate.txt"
> Query performance on S3 files
> -----------------------------
>
> Key: DRILL-6814
> URL: https://issues.apache.org/jira/browse/DRILL-6814
> Project: Apache Drill
> Issue Type: Improvement
> Components: Storage - Other
> Affects Versions: 1.14.0
> Environment: Amazon EC2 instances-
> 4 Linux Redhat machines -version 7.5
> RAM- 32GB
> Reporter: Ashish Shukla
> Assignee: Denys Ordynskiy
> Priority: Major
>
> I have installed 4 Node drill cluster on Amazon EC2 and trying to execute a
> simple count on one Amazon S3 file. File type is CSV and size is approx- 14GB.
> The query returns expected count after the execution of approx 30 minutes.
> If we keep the same file in hdfs or create a table in postgres, execution
> time is relatively very less (approx 2-3 minutes).
> Is it normal behavior or something can be done for S3 files to make
> execution time comparable ?
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
