[ 
https://issues.apache.org/jira/browse/SOLR-475?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12650071#action_12650071
 ] 

Yonik Seeley commented on SOLR-475:
-----------------------------------

Some further results on a bigger index to show some practical limits.
This table (JIRA markup format) shows the performance and memory 
characteristics of facet requests on a 50M document index, for different fields 
and different numbers of documents being counted in the base query.

|| ||f10_100_t||f100_10_t||f1000_5_t||f10000_5_t||f100000_5_t||f100000_10_t
|field inversion time (sec)|    17.2|   17.9|   69.4|   87.8|   133.6|  388.0
|inverted field size (MB)|      68.1|   629.6|  416.9|  479.0|  589.9|  807.4
|1000 docs facet time (ms)|     7|      20|     13      |13     |16     |17
|10,000 docs|   55|     428     |22|    23      |29|    28
|100,000 docs|  54      |421|   35      |36     |46     |56
|1,000,000 docs|        55|     431     |149    |155|   249     |307
|10,000,000 docs|       54| 434|        625|    625     |1183|  1620

The "profile" of the faceted field is encoded in it's name.  For example, the 
field f1000_5_t has 1000 unique values across the whole index and between 0 and 
5 values per document.  It took 35 ms to facet on this field when the base 
query matched 100,000 documents.


Test Hardware: Commodity PC
 Processor: AMD Athlon 64 X2 5000+ (2.6GHz dual core)
Hard Drive: Western Digital Caviar GP WD5000AACS 500GB 5400 to 7200 RPM SATA 
3.0Gb/s
Memory: 8GB DDR2 800 SDRAM (PC2 6400)
Operating System: Linux - Ubuntu 8.04 desktop, 64 bit version (x86_64)
Java VM: Sun Java6 (1.6.0_05) 64 bit hotspot (x86_64)



> multi-valued faceting via un-inverted field
> -------------------------------------------
>
>                 Key: SOLR-475
>                 URL: https://issues.apache.org/jira/browse/SOLR-475
>             Project: Solr
>          Issue Type: New Feature
>            Reporter: Yonik Seeley
>         Attachments: facet_performance.html, UnInvertedField.java, 
> UnInvertedField.java
>
>
> Facet multi-valued fields via a counting method (like the FieldCache method) 
> on an un-inverted representation of the field.  For each doc, look at it's 
> terms and increment a count for that term.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

Reply via email to