diff options
author | Sital Kedia <skedia@fb.com> | 2016-08-04 14:54:38 -0700 |
---|---|---|
committer | Josh Rosen <joshrosen@databricks.com> | 2016-08-04 14:54:38 -0700 |
commit | 9c15d079df2418a1412269a702f3a7861daee61c (patch) | |
tree | 713b294ebc99767ad3143df8815bbe2014f2cd3e /docs | |
parent | 0e2e5d7d0b42226c61c3200fd63d2831c558519d (diff) | |
download | spark-9c15d079df2418a1412269a702f3a7861daee61c.tar.gz spark-9c15d079df2418a1412269a702f3a7861daee61c.tar.bz2 spark-9c15d079df2418a1412269a702f3a7861daee61c.zip |
[SPARK-15074][SHUFFLE] Cache shuffle index file to speedup shuffle fetch
## What changes were proposed in this pull request?
Shuffle fetch on large intermediate dataset is slow because the shuffle service open/close the index file for each shuffle fetch. This change introduces a cache for the index information so that we can avoid accessing the index files for each block fetch
## How was this patch tested?
Tested by running a job on the cluster and the shuffle read time was reduced by 50%.
Author: Sital Kedia <skedia@fb.com>
Closes #12944 from sitalkedia/shuffle_service.
Diffstat (limited to 'docs')
-rw-r--r-- | docs/configuration.md | 7 |
1 files changed, 7 insertions, 0 deletions
diff --git a/docs/configuration.md b/docs/configuration.md index bf10b24819..cc6b2b6470 100644 --- a/docs/configuration.md +++ b/docs/configuration.md @@ -522,6 +522,13 @@ Apart from these, the following properties are also available, and may be useful </td> </tr> <tr> + <td><code>spark.shuffle.service.index.cache.entries</code></td> + <td>1024</td> + <td> + Max number of entries to keep in the index cache of the shuffle service. + </td> +</tr> +<tr> <td><code>spark.shuffle.sort.bypassMergeThreshold</code></td> <td>200</td> <td> |