Re: 3GB RAM usage by Oracle

From: Howard J. Rogers <hjr_at_dizwell.com>
Date: Tue, 31 Aug 2004 10:04:45 +1000
Message-ID: <4133c0b6$0$14486$afc38c87@news.optusnet.com.au>

Joel Garry wrote:

> don_at_burleson.cc (Don Burleson) wrote in message
> news:<998d28f7.0408290431.6e55433f_at_posting.google.com>...

>>
>> 3 ? RAM is an IMPORTANT exception to Moore's Law because it DOES NOT
>> increase in speed every year. (Hint: this is VERY important point)

> 
> Moore's Law is transistor density, not speed.
> http://www.intel.com/research/silicon/mooreslaw.htm
> 

>>
>> 4 ? Even a well-tuned Oracle databases has a bottleneck.
>>
>> 5 - Management often ignores the "big picture" and makes decisions
>> based solely on CYA, total cost, performance benefit and total
>> database downtime.
>>
>> Ergo:
>>
>> 1 - Throwing hardware at an Oracle problem is a legitimate management
>> decision, and in some cases a cost-effective one.
>>
>> 2 - Faster CPU will speed-up ANY CPU-Bound Oracle database.
>>
>> 3 - High-speed SSD will speed up ANY I/O-bound Oracle database.
>>
>> 4 - Faster network will speed-up ANY "SQL*Net" bound Oracle database.

> 
> Oddly enough, I've seen cases where these are not true.  For example,
> speeding up the cpu leading to more wait state _conflicts_ for I/O,
> which cascades into a slower overall response time noticed by users.
> Kinda like siphoning more gas outa Ol' Yeller with a smooth flow
> rather than it going blorp-blorp.

I agree with you Joel.

It is not obvious that buying a faster CPU will speed up any CPU-bound Oracle database if all it does is to eliminate a CPU bottleneck. Quite possibly, that simply then exposes the fact that huge amounts of I/O are taking place, and the hard disk becomes the next bottleneck. So you buy more hard disks, and discover that the LVM it uses chews up CPU cycles... and so on.

An over-simplified picture, of course.

But not quite as over-simplified as saying, without qualification, that extra resource WILL speed up a database. That's a ludicrous proposition.

Regards
HJR Received on Mon Aug 30 2004 - 19:04:45 CDT