You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@solr.apache.org by "Mark Robert Miller (Jira)" <ji...@apache.org> on 2021/09/29 16:05:00 UTC

[jira] [Commented] (SOLR-15666) Heavily reduce faceting ram usage for many cases.

    [ https://issues.apache.org/jira/browse/SOLR-15666?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17422219#comment-17422219 ] 

Mark Robert Miller commented on SOLR-15666:
-------------------------------------------

Basic before and after gc picture for the very narrow faceting benchmark currently in:

Before
{noformat}
Benchmark                                                (docCount)  (facetCard)  (facetCard2)   (fm)  (nodeCount)  (numReplicas)  (numShards) Mode  Cnt          Score           Error   Units
JsonFaceting.jsonFacet                                       500000        15000          3000  smart            2              1            4thrpt    5         13.075 ±         0.593   ops/s
JsonFaceting.jsonFacet:·async                                500000        15000          3000  smart            2              1            4thrpt                 NaN                     ---
JsonFaceting.jsonFacet:·gc.alloc.rate                        500000        15000          3000  smart            2              1            4thrpt    5       1233.493 ±      2651.035  MB/sec
JsonFaceting.jsonFacet:·gc.alloc.rate.norm                   500000        15000          3000  smart            2              1            4thrpt    5  103587267.304 ± 222574550.680    B/op
JsonFaceting.jsonFacet:·gc.churn.G1_Eden_Space               500000        15000          3000  smart            2              1            4thrpt    5       1556.852 ±       368.148  MB/sec
JsonFaceting.jsonFacet:·gc.churn.G1_Eden_Space.norm          500000        15000          3000  smart            2              1            4thrpt    5  131303411.889 ±  30279661.952    B/op
JsonFaceting.jsonFacet:·gc.churn.G1_Old_Gen                  500000        15000          3000  smart            2              1            4thrpt    5          0.330 ±         0.455  MB/sec
JsonFaceting.jsonFacet:·gc.churn.G1_Old_Gen.norm             500000        15000          3000  smart            2              1            4thrpt    5      27858.041 ±     38880.588    B/op
JsonFaceting.jsonFacet:·gc.churn.G1_Survivor_Space           500000        15000          3000  smart            2              1            4thrpt    5          2.988 ±         3.294  MB/sec
JsonFaceting.jsonFacet:·gc.churn.G1_Survivor_Space.norm      500000        15000          3000  smart            2              1            4thrpt    5     252175.421 ±    277771.041    B/op
JsonFaceting.jsonFacet:·gc.count                             500000        15000          3000  smart            2              1            4thrpt    5         41.000                  counts
JsonFaceting.jsonFacet:·gc.time                              500000        15000          3000  smart            2              1            4thrpt    5        218.000                                                 500000        15000          3000  smart            2              1            4thrpt    5        79.000{noformat}
After
{noformat}
Benchmark                                                (docCount)  (facetCard)  (facetCard2)   (fm)  (nodeCount)  (numReplicas)  (numShards) Mode  Cnt         Score          Error   Units
JsonFaceting.jsonFacet                                       500000        15000          3000  smart            2              1            4thrpt    5        15.311 ±        0.385   ops/s
JsonFaceting.jsonFacet:·async                                500000        15000          3000  smart            2              1            4thrpt                NaN                    ---
JsonFaceting.jsonFacet:·gc.alloc.rate                        500000        15000          3000  smart            2              1            4thrpt    5       357.791 ±      765.257  MB/sec
JsonFaceting.jsonFacet:·gc.alloc.rate.norm                   500000        15000          3000  smart            2              1            4thrpt    5  25560743.208 ± 54651776.034    B/op
JsonFaceting.jsonFacet:·gc.churn.G1_Eden_Space               500000        15000          3000  smart            2              1            4thrpt    5       449.762 ±       27.717  MB/sec
JsonFaceting.jsonFacet:·gc.churn.G1_Eden_Space.norm          500000        15000          3000  smart            2              1            4thrpt    5  32363639.760 ±  2609924.751    B/op
JsonFaceting.jsonFacet:·gc.churn.G1_Old_Gen                  500000        15000          3000  smart            2              1            4thrpt    5         0.061 ±        0.291  MB/sec
JsonFaceting.jsonFacet:·gc.churn.G1_Old_Gen.norm             500000        15000          3000  smart            2              1            4thrpt    5      4359.291 ±    20832.751    B/op
JsonFaceting.jsonFacet:·gc.churn.G1_Survivor_Space           500000        15000          3000  smart            2              1            4thrpt    5         0.815 ±        4.890  MB/sec
JsonFaceting.jsonFacet:·gc.churn.G1_Survivor_Space.norm      500000        15000          3000  smart            2              1            4thrpt    5     58042.222 ±   345702.114    B/op
JsonFaceting.jsonFacet:·gc.count                             500000        15000          3000  smart            2              1            4thrpt    5        15.000                 counts
JsonFaceting.jsonFacet:·gc.time                              500000        15000          3000  smart            2              1            4thrpt    5        79.000{noformat}

> Heavily reduce faceting ram usage for many cases.
> -------------------------------------------------
>
>                 Key: SOLR-15666
>                 URL: https://issues.apache.org/jira/browse/SOLR-15666
>             Project: Solr
>          Issue Type: Sub-task
>            Reporter: Mark Robert Miller
>            Assignee: Mark Robert Miller
>            Priority: Major
>
> Spinning this out of various efforts from the parallel segments search branches.
> Would require a bit more tuning and a much more extensive benchmark verification effort.
> This has performance impact as well given gc costs of faceting.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@solr.apache.org
For additional commands, e-mail: issues-help@solr.apache.org