[
https://issues.apache.org/jira/browse/HADOOP-6147?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12730460#action_12730460
]
Jay Booth commented on HADOOP-6147:
-----------------------------------
It's more efficient raw storage and you can mount it however you want -- from
what I understand (only having messed with it very little), this would be
something in how you set up your AMI, i.e. you set it up with a couple EBS
stores that are striped RAID 0 and it will outperform the 'ephemeral' storage
that you get for free with the instance. Major gain is that since you pay by
gigabyte, you can do 4 5GB stores for the same price as 1 20GB store, so you
can go either RAID 0 or RAID 5 depending on how you feel about node failure,
and get faster IO for the same (or almost the same) cost per GB. After that
setup's done, though, you just mount it to /mnt or /opt or wherever and use it
like you would any other device, so I don't see much need to hadoop specific
code. If I had to guess, what cloudera's distributing is a custom AMI, not any
significant modification to Hadoop. Any of the above could be way off, though.
> Hadoop Filesystem Plugin for Amazon Elastic Block Storage (EBS)
> ---------------------------------------------------------------
>
> Key: HADOOP-6147
> URL: https://issues.apache.org/jira/browse/HADOOP-6147
> Project: Hadoop Common
> Issue Type: New Feature
> Reporter: Sanjay Radia
>
> Amazon has introduced a new storage mechanism called EBS. Supposedly it is a
> lot more efficient than S3 for Hadoop.
> A filesystem plugin for hadoop would we be very useful for the community.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
