Re: tmpfs and postgres memory

From: Greg Spiegelberg <gspiegelberg(at)gmail(dot)com>
To: Anj Adu <fotographs(at)gmail(dot)com>
Cc: pgsql-performance(at)postgresql(dot)org
Subject: Re: tmpfs and postgres memory
Date: 2010-04-27 03:59:50
Message-ID: w2z22723571004262059n36f065ai5ff22e08a6920a2c@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-performance

On Mon, Apr 26, 2010 at 5:24 PM, Anj Adu <fotographs(at)gmail(dot)com> wrote:
> I have a 16G box and tmpfs is configured to use 8G for tmpfs .
>
> Is a lot of memory being wasted that can be used for Postgres ? (I am
> not seeing any performance issues, but I am not clear how Linux uses
> the tmpfs and how Postgres would be affected by the reduction in
> memory)

Like Solaris, tmpfs is from swap and swap is both memory and disk so
there is no guarantee when you're using it that it will be the fast
memory based file system you're looking for.

What you may be wanting is ramfs. Unlike tmpfs, it is 100% memory.
Another difference is though you may mount a ramfs file system
specifying a size, no real size is enforced. If you have 2GB of
memory and attempt to copy 2GB of files to a ramfs mount point the
system will do it until all the space, i.e. memory, is gone.

Both tmpfs and ramfs come with a price, that is at the flick of a
switch, loss of power or other cause that resets or reboots the system
all data is lost. That reason doesn't necessarily mean you can't use
a memory based file system it just limits it's applications.

Personally, I'd find a way to tell PostgreSQL about the memory before
toying with tmpfs or ramfs but I'm sure our applications are
different.

-Greg

In response to

Browse pgsql-performance by date

  From Date Subject
Next Message Robert Haas 2010-04-28 00:46:36 Re: Optimization idea
Previous Message Anj Adu 2010-04-26 23:24:14 tmpfs and postgres memory