[ 
https://issues.apache.org/jira/browse/NUTCH-506?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12513044
 ] 

Doğacan Güney commented on NUTCH-506:
-------------------------------------

If there are no objections, I am going to commit this one.

Just to get more comments, here is a break-down of what this patch does:

* Remove all compression code from nutch. This means no more 
writeCompressedString-s or writeCompressedStringArrays, also no more 
CompressedWritable-s. All changes are done in a backward compatible manner. 
Also after this change, Content's version is -1 and new changes should 
*decrease* that number. See NUTCH-392 for more details.

* Respect io.seqfile.compression.type setting for all structures except 
ParseText. ParseText is always compressed as RECORD. Also for some reason 
crawl_generate is not compressed.

Why are we doing this? Because hadoop can efficiently (both in space and in 
time) compress these structures for us. I have done some tests with different 
compression settings in NUTCH-392 and BLOCK compression really does a 
difference. I think for a large enough crawl, overall space savings will be 
around %20 - %40. Note that this is basically for free (there may even be a 
small performance gain) if you are using hadoop's native libraries.

> Nutch should delegate compression to Hadoop
> -------------------------------------------
>
>                 Key: NUTCH-506
>                 URL: https://issues.apache.org/jira/browse/NUTCH-506
>             Project: Nutch
>          Issue Type: Improvement
>            Reporter: Doğacan Güney
>             Fix For: 1.0.0
>
>         Attachments: compress.patch, NUTCH-506.patch
>
>
> Some data structures within nutch (such as Content, ParseText) handle their 
> own compression. We should delegate all compressions to Hadoop. 
> Also, nutch should respect io.seqfile.compression.type setting. Currently 
> even if io.seqfile.compression.type is BLOCK or RECORD, nutch overrides it 
> for some structures and sets it to NONE (However, IMO, ParseText should 
> always be compressed as RECORD because of performance reasons).

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


-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
_______________________________________________
Nutch-developers mailing list
Nutch-developers@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nutch-developers

Reply via email to