[
https://issues.apache.org/jira/browse/HBASE-2315?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12844731#action_12844731
]
Flavio Paiva Junqueira commented on HBASE-2315:
-----------------------------------------------
Thanks for all comments and questions so far. Let me try to answer all in a
batch:
Stack: We are certainly interested in exploring, and learning your requirements
would certainly help to guide our exploration. On making BK transparent to the
user installing BK, I don't have any reason to think that it would be difficult
to achieve. The script I use to start a bookie is very simple and we'll soon
have the feature I was mentioning above of a bookie registering automatically
with ZooKeeper. In any case, I made sure that it is possible to use BK
optionally, so hdfs is still used by default with the current patch.
Ryan, Andrew: The number of nodes going from 20 to a 100 is not a problem; We
have tested BK with a few thousand concurrent ledgers. One important aspect to
pay attention to is the length of writes. In our experience, BK is more
efficient when we batch small requests. If it is possible to batch small
requests, then I don't see an issue with obtaining the throughput figures
you're asking for. In fact, with a previous version of BK, we were able to get
70kwrites/s for writes of 128 bytes.
Todd: There is some documentation available on the ZooKeeper site, and we have
added more for the next release (3.3.0). You can get the new documentation from
trunk, but I'll also add a pdf here for your convenience.
> BookKeeper for write-ahead logging
> ----------------------------------
>
> Key: HBASE-2315
> URL: https://issues.apache.org/jira/browse/HBASE-2315
> Project: Hadoop HBase
> Issue Type: New Feature
> Components: regionserver
> Reporter: Flavio Paiva Junqueira
> Attachments: HBASE-2315.patch, zookeeper-dev-bookkeeper.jar
>
>
> BookKeeper, a contrib of the ZooKeeper project, is a fault tolerant and high
> throughput write-ahead logging service. This issue provides an implementation
> of write-ahead logging for hbase using BookKeeper. Apart from expected
> throughput improvements, BookKeeper also has stronger durability guarantees
> compared to the implementation currently used by hbase.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
