You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Yanfei Lei (Jira)" <ji...@apache.org> on 2022/11/07 08:50:00 UTC
[jira] [Commented] (FLINK-28539) Enable CompactionDynamicLevelBytes in FLASH_SSD_OPTIMIZED
[ https://issues.apache.org/jira/browse/FLINK-28539?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17629684#comment-17629684 ]
Yanfei Lei commented on FLINK-28539:
------------------------------------
Since 1) We can enable CompactionDynamicLevelBytes for SSD by configuring "RocksDBConfigurableOptions.USE_DYNAMIC_LEVEL_SIZE: true" manually.
2)From [RocksDB wiki | [https://github.com/facebook/rocksdb/wiki/Tuning-RocksDB-on-Spinning-Disks]] , it is recommended to set set level_compaction_dynamic_level_bytes for spinning disk
I am going to close this ticket.
> Enable CompactionDynamicLevelBytes in FLASH_SSD_OPTIMIZED
> ---------------------------------------------------------
>
> Key: FLINK-28539
> URL: https://issues.apache.org/jira/browse/FLINK-28539
> Project: Flink
> Issue Type: Improvement
> Components: Runtime / State Backends
> Reporter: Usamah Jassat
> Priority: Minor
>
> Investigating the RocksDB predefined options I see that `setLevelCompactionDynamicLevelBytes` is set for SPINNING_DISK options but not FLASH_SSD_OPTIMIZED.
>
> From my research it looks like this change would improve the Space Amplification of RocksDB [1] (which can also lead to a trade-off from read/write amplification [2]). It makes sense to me that this feature should be enabled for SSD's as they tend to have less space compared to their HDD counterparts.
> There is also an argument to be made to also disable it for SPINNING_DISK options as it could give increased read/write performance [2]
> [1] [http://rocksdb.org/blog/2015/07/23/dynamic-level.html]
> [2] [https://github.com/EighteenZi/rocksdb_wiki/blob/master/RocksDB-Tuning-Guide.md#amplification-factors]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)