Message-ID: <4797AC4E.6040007@emolecules.com>
Date: Wed, 23 Jan 2008 13:06:22 -0800
From: Craig James <craig_james@emolecules.com>
User-Agent: Thunderbird 2.0.0.9 (Macintosh/20071031)
MIME-Version: 1.0
To: Guy Rouillier <guyr-ml1@burntmail.com>, pgsql-performance@postgresql.org
Subject: Re: Making the most of memory?
References: <4796BE84.9030803@centriccrm.com>	
	<dcc563d10801222120q78e390dbj2b884eca729c2975@mail.gmail.com>	
	<B30794E6-CD2A-43C1-B431-E24B56FBFB1D@concursive.com>
	<dcc563d10801231044p1dba892r9b8bf1269395f068@mail.gmail.com>
	<47979C15.3040406@burntmail.com>
In-Reply-To: <47979C15.3040406@burntmail.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Guy Rouillier wrote:
> Scott Marlowe wrote:
>> I assume you're talking about solid state drives?  They have their
>> uses, but for most use cases, having plenty of RAM in your server will
>> be a better way to spend your money.  For certain high throughput,
>> relatively small databases (i.e. transactional work) the SSD can be
>> quite useful.
> 
> Unless somebody has changes some physics recently, I'm not understanding 
> the recent discussions of SSD in the general press.  Flash has a limited 
> number of writes before it becomes unreliable.  On good quality consumer 
> grade, that's about 300,000 writes, while on industrial grade it's about 
> 10 times that.  That's fine for mp3 players and cameras; even 
> professional photographers probably won't rewrite the same spot on a 
> flash card that many times in a lifetime.  But for database 
> applications, 300,000 writes is trivial. 3 million will go a lot longer, 
> but in non-archival applications, I imagine even that mark won't take 
> but a year or two to surpass.

One trick they use is to remap the physical Flash RAM to different logical addresses.  Typical apps update a small percentage of the data frequently, and the rest of the data rarely or never.  By shuffling the physical Flash RAM around, the media lasts a lot longer than a simple analysis might indicate.

Craig