Master/Slave, DB separation or just spend $$$? - Mailing list pgsql-performance

From Kelvin Quee
Subject Master/Slave, DB separation or just spend $$$?
Date
Msg-id 4e9464f90907212047p799db680o650b4a9754547d63@mail.gmail.com
Whole thread Raw
Responses Re: Master/Slave, DB separation or just spend $$$?
Re: Master/Slave, DB separation or just spend $$$?
List pgsql-performance
Hi Performance Wizards!

I need advice on this.

I have a db which is being constantly updated and queried by a few
computers. We are doing datamining. The machine is running on a
moderately powered machine and processors constantly hit 90%.

At the same time, we need to present these data on a web interface.
The performance for the web interface is now very sluggish as most of
the power is occupied by the mining process.

I have thought of a few ways out of this -

1) Buy a mega powered machine (temporal solution, quick fix)
2) Do a master-slave configuration
3) Separate the DB into 2 - One for pure mining purposes, the other
purely for web serving

For (2), I do not know if it will be very effective since the master
will probably have many changes at any moment. I do not understand how
the changes will be propagated from the master to the slave without
impacting the slave's performance. Anyone with more experience here?

(3) seems ideal but is a very very painful solution!

We can possibly use a message queue system but again I am not familiar
with MQ. Will need to do more research.

If you were me, how would you solve this problem?

Thanks!


Kelvin Quee
+65 9177 3635

pgsql-performance by date:

Previous
From: Oleg Bartunov
Date:
Subject: Re: Full text search with ORDER BY performance issue
Next
From: "Victor de Buen (Bayes)"
Date:
Subject: Atomic access to large arrays