[jira] Issue Comment Edited: (CASSANDRA-2045) Simplify HH to decrease read load when nodes come back

["Chris Goffinet (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/CASSANDRA-2045?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12986087#action_12986087 ] 

Chris Goffinet edited comment on CASSANDRA-2045 at 1/24/11 6:20 PM:
--------------------------------------------------------------------

I disagree, storing a pointer is not almost free. The trade off you make is read performance when nodes come back up. At the moment read performance could very well cause cascading failures too. You trade storage vs I/O.  Nodes down for long periods of time, would have a bigger impact on the nodes trying to send HH data.

Can you clarify, aren't we sending the entire row as well? We might be doing paging but if I modify 1 column, and it has 1M columns, 1M + 1 columns still get sent? I would agree to having it be tunable, but I prefer consistency of performance over storage when it comes to I/O. 

We brought this up because we actually saw this in our live clusters.

      was (Author: lenn0x):
    No, storing a pointer is not almost free. The trade off you make is read performance when nodes come back up. At the moment read performance could very well cause cascading failures too. You trade storage vs I/O.  Nodes down for long periods of time, would have a bigger impact on the nodes trying to send HH data.

Aren't we sending the entire row as well? We might be doing paging but if I modify 1 column, and it has 1M columns, 1M + 1 columns still get sent? I would agree to having it be tunable, but I prefer performance consistency over storage when it comes to I/O.
  
> Simplify HH to decrease read load when nodes come back
> ------------------------------------------------------
>
>                 Key: CASSANDRA-2045
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-2045
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: Chris Goffinet
>             Fix For: 0.7.2
>
>
> Currently when HH is enabled, hints are stored, and when a node comes back, we begin sending that node data. We do a lookup on the local node for the row to send. To help reduce read load (if a node is offline for long period of time) we should store the data we want forward the node locally instead. We wouldn't have to do any lookups, just take byte[] and send to the destination.

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