You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@nutch.apache.org by "Julien Nioche (JIRA)" <ji...@apache.org> on 2009/08/28 16:58:59 UTC

[jira] Commented: (NUTCH-702) Lazy Instanciation of Metadata in CrawlDatum

    [ https://issues.apache.org/jira/browse/NUTCH-702?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12748840#action_12748840 ] 

Julien Nioche commented on NUTCH-702:
-------------------------------------

There have been quite a few related questions on the mailing-lists since releasing the patch. As pointed out by Andrzej, the best way of avoiding memory exceptions is to limit the number of URLs per inlink (see nutch-default.xml), however this patch allows to set a higher value for this parameter as it reduces the memory footprint. 

The main advantage of this patch is that it speeds up all phases of a crawl involving  the crawlDB. As an illustration I did a bit of comparison between the default version and the patched one, the average results being : 

Injection (1M URLs)
original : 98 secs
patched : 65 secs

generation : 
original : 37 secs
patched : 37 secs

stats: 
original : 13 secs
patched : 19 secs

update: 
original : 40 secs
patched : 19 secs


I find a bit surprising that the patched version is not faster on the generation (any idea?). On the other phases it seems to be 50% faster, except for the updating where it is twice as fast.

J.


> Lazy Instanciation of Metadata in CrawlDatum
> --------------------------------------------
>
>                 Key: NUTCH-702
>                 URL: https://issues.apache.org/jira/browse/NUTCH-702
>             Project: Nutch
>          Issue Type: Improvement
>    Affects Versions: 1.0.0
>            Reporter: Julien Nioche
>         Attachments: NUTCH-702.patch, NUTCH-702.patch.v2, NUTCH-702v3.patch
>
>
> CrawlDatum systematically instanciates its metadata, which is quite wasteful especially in the case of CrawlDBReducer when it generates a new CrawlDatum for each incoming link before storing it in a List.
> Initial testing on the lazy instanciation shows an improvement in both speed and memory consumption. I will generate a patch for it ASAP

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