You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@hbase.apache.org by Rohit Kelkar <ro...@gmail.com> on 2012/08/24 07:01:50 UTC
restriction on the number of tables in hbase and its impact on performance
I have asked this question on stackoverflow -
http://stackoverflow.com/questions/12066856/restriction-on-the-number-of-tables-in-hbase-and-its-impact-on-performance
Also asking the same on this list --
Our hbase schema in production has 5 tables. We have N clients where
in only 10% of the clients are active at any given instant. So for me
it looks like a waste of resources to keep the data of remaining 90%
clients active. I was thinking of creating 5 tables per client so that
I can keep the active client's tables enabled and the remaining
client's tables disabled. From what I have read if we exceed 1000
regions per region server then performance starts degrading. But I am
sure not to hit that limit. My questions
If I disable a set of tables then does it mean that I am putting less
load on hbase?
Does this seem like a sound strategy overall?
- Rohit Kelkar
Re: restriction on the number of tables in hbase and its impact on performance
Posted by Kevin O'dell <ke...@cloudera.com>.
Rohit,
This is an interesting question, but it sounds like overkill. I would
not worry about having tables up that aren't active. If you keep your
active region count down and your memory footprint reasonable <16GB heap
you should be fine.
On Fri, Aug 24, 2012 at 1:01 AM, Rohit Kelkar <ro...@gmail.com> wrote:
> I have asked this question on stackoverflow -
>
> http://stackoverflow.com/questions/12066856/restriction-on-the-number-of-tables-in-hbase-and-its-impact-on-performance
>
> Also asking the same on this list --
>
> Our hbase schema in production has 5 tables. We have N clients where
> in only 10% of the clients are active at any given instant. So for me
> it looks like a waste of resources to keep the data of remaining 90%
> clients active. I was thinking of creating 5 tables per client so that
> I can keep the active client's tables enabled and the remaining
> client's tables disabled. From what I have read if we exceed 1000
> regions per region server then performance starts degrading. But I am
> sure not to hit that limit. My questions
>
> If I disable a set of tables then does it mean that I am putting less
> load on hbase?
> Does this seem like a sound strategy overall?
>
> - Rohit Kelkar
>
--
Kevin O'Dell
Customer Operations Engineer, Cloudera