region number, region merge and performance

[Libo Yu <yu...@hotmail.com>]

Hi all,

Because we have used the current time stamp in the key, as time goes on, we have 
more than more regions. After TTL expires, many of regions are empty or almost 
empty. So currently, our region servers have many regions on them but only a small
number of them are active (receive read/write requests). 
So my questions are:

1 How does the number of region impact the performance? What I can think of is that
MSLAB is enabled, that will waste a lot memory.

2 I did some search. Hbase has utilities for merge regions and some people suggest to 
recreate the table and reload the table. But if I want to merge regions for all the entities 
on my cluster automatically,  what is the best way to do it?

Thanks,

Linlin



 		 	   		  

Re: region number, region merge and performance

[Ted Yu <yu...@gmail.com>]

For online merge of regions, please take a look at HBASE-7403

Cheers


On Fri, Mar 14, 2014 at 6:18 AM, Libo Yu <yu...@hotmail.com> wrote:

> Hi all,
>
> Because we have used the current time stamp in the key, as time goes on,
> we have
> more than more regions. After TTL expires, many of regions are empty or
> almost
> empty. So currently, our region servers have many regions on them but only
> a small
> number of them are active (receive read/write requests).
> So my questions are:
>
> 1 How does the number of region impact the performance? What I can think
> of is that
> MSLAB is enabled, that will waste a lot memory.
>
> 2 I did some search. Hbase has utilities for merge regions and some people
> suggest to
> recreate the table and reload the table. But if I want to merge regions
> for all the entities
> on my cluster automatically,  what is the best way to do it?
>
> Thanks,
>
> Linlin
>
>
>
>