is postgres a good solution for billion record data

Started by shahrzad khorramiover 16 years ago3 messagesgeneral
Jump to latest
#1shahrzad khorrami
shahrzad.khorrami@gmail.com

is postgres a good solution for billion record data, think of 300kb data
insert into db at each minutes, I'm coding with php
what do you recommend to manage these data?

--
Shahrzad Khorrami

#2Scott Marlowe
scott.marlowe@gmail.com
In reply to: shahrzad khorrami (#1)
Re: is postgres a good solution for billion record data

On Sat, Oct 24, 2009 at 7:32 AM, shahrzad khorrami
<shahrzad.khorrami@gmail.com> wrote:

is postgres a good solution for billion record data, think of 300kb data
insert into db at each minutes, I'm coding with php
what do you recommend to manage these data?

You'll want a server with LOTS of hard drives spinning under it. Fast
RAID controller with battery backed RAM. Inserting the data is no
problem. 300kb a minute is nothing. My stats machine that handles
about 2.5M rows a day during the week is inserting in the megabytes
per second (it's also the search database so there's the indexer wtih
16 threads hitting it). The stats part of the load is miniscule until
you start retrieving large chunks of data, then it's mostly sequential
reads in the 100+Megs a second.

The more drives and the better the RAID controller you throw at the
problem the better performance you'll get. For the price of one
oracle license for one core, you can build a damned find pgsql server
or pair of servers.

#3Scott Marlowe
scott.marlowe@gmail.com
In reply to: Scott Marlowe (#2)
Re: is postgres a good solution for billion record data

On Sat, Oct 24, 2009 at 2:43 PM, Scott Marlowe <scott.marlowe@gmail.com> wrote:

On Sat, Oct 24, 2009 at 7:32 AM, shahrzad khorrami
<shahrzad.khorrami@gmail.com> wrote:

is postgres a good solution for billion record data, think of 300kb data
insert into db at each minutes, I'm coding with php
what do you recommend to manage these data?

You'll want a server with LOTS of hard drives spinning under it.  Fast
RAID controller with battery backed RAM.  Inserting the data is no
problem. 300kb a minute is nothing.  My stats machine that handles
about 2.5M rows a day during the week is inserting in the megabytes
per second (it's also the search database so there's the indexer wtih
16 threads hitting it).  The stats part of the load is miniscule until
you start retrieving large chunks of data, then it's mostly sequential
reads in the 100+Megs a second.

The more drives and the better the RAID controller you throw at the
problem the better performance you'll get.  For the price of one
oracle license for one core, you can build a damned find pgsql server
or pair of servers.

Quick reference, you get one of these:

http://www.aberdeeninc.com/abcatg/Stirling-X888.htm

with dual 2.26GHz Nehalem CPUs, 48 Gigs ram, and 48 73K 15kRPM Seagate
barracudas for around $20,000. That's the same cost for a single
oracle license for one CPU. That's way overkill for what you're
talking about doing. A machine with 8 or 16 disks could easily handle
the load you're talking about.