phoenix-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Puneet Kumar Ojha <puneet.ku...@pubmatic.com>
Subject RE: Salt buckets optimization
Date Fri, 27 Feb 2015 06:34:42 GMT


From: Puneet Kumar Ojha [mailto:puneet.kumar@pubmatic.com]
Sent: Friday, February 27, 2015 12:00 PM
To: user@phoenix.apache.org
Subject: RE: Salt buckets optimization

Yes you are correct.
There will be two level of aggregation . One in the each region and then aggregating records
of each region.
Fewer regions will help in less records at client side aggregation.

Thanks
Puneet.



From: Krishna [mailto:research800@gmail.com]
Sent: Friday, February 27, 2015 6:59 AM
To: user@phoenix.apache.org<mailto:user@phoenix.apache.org>
Subject: Re: Salt buckets optimization

Puneet, In addition to creating an initial number of regions, it appears, salt buckets is
also used for determining the # of threads (?) while doing skip scan. Explain plan shows something
like "client 60-way skip scan...". So, having a higher number of salt buckets should, theoretically,
help query performance. Could you explain what kind of issues could pop-up having 150 to 200+
salt bytes?



On Thu, Feb 26, 2015 at 2:57 AM, Puneet Kumar Ojha <puneet.kumar@pubmatic.com<mailto:puneet.kumar@pubmatic.com>>
wrote:
Yes you can have more salt buckets than no of regions servers. Salt buckets specififed will
create thoe many number of regions.

Depending upon what regions size is optimal –(dependent on table data) you need to opt the
salt bucket numbers.


From: Vamsi Krishna [mailto:vamsi.attluri@gmail.com<mailto:vamsi.attluri@gmail.com>]
Sent: Thursday, February 26, 2015 11:25 AM
To: user@phoenix.apache.org<mailto:user@phoenix.apache.org>
Subject: Re: Salt buckets optimization

Can we have more no of salt buckets than the no of region servers?
According to Phoenix documentation, I've read "For optimal performance, number of salt buckets
should match number of region servers".
http://phoenix.apache.org/performance.html

In our case, we are evaluating on a small cluster with two region servers and we are observing
better performance with 10 salt buckets compared to 2 salt buckets and 20 salt buckets.

Could someone please explain what's happening behind the scenes?

Thanks,
Vamsi Attluri.

On Wed, Feb 25, 2015 at 6:56 PM, Puneet Kumar Ojha <puneet.kumar@pubmatic.com<mailto:puneet.kumar@pubmatic.com>>
wrote:

For big tables keep the salt bucket high, generally around 60-90.

Smaller or join tables should have salt buckets as minimal may be 1-4.





Thanks





------ Original message------

From: Krishna

Date: Thu, Feb 26, 2015 07:42

To: dev@phoenix.apache.org<mailto:dev@phoenix.apache.org>;user@phoenix.apache.org<mailto:user@phoenix.apache.org>;

Subject:Salt buckets optimization


Are there any recommendations for estimating and optimizing salt buckets during table creation
time? What, if any, are the cons of having high number (200+) of salt buckets? Is it possible
to update salt buckets after table is created?

Thanks



Mime
View raw message