RE: Oracle RAC cost justification?

It seems everyone keeps dodging the question of measurement.

Is the sum of "inconvenience" to the business process of a robustly configured SMP
greater or less than the sum of "inconvenience" to the business process of a robustly
configured RAC system?

First, you have to define what is "inconvenient" in your particular business situation.

For some businesses (in fact probably yours, since you have an identified time of requiring
high availibility) there is little or no inconvenience to outages for preventive maintenance
and upgrades. (Call this polar case A for reference.)

For other businesses any outage at any time is a problem. (Case B).

Second, what do I mean by "robustly configured?" That means no stupidly configured single
points of failure, and only considering failure points that a reasonably smart and
experienced compulsive obsessive sysadmin/dba team would not preempt in configuration.

Now again, this is really a question for a measurement and/or simulation, but based on
what I've seen:

In Case A you're far more likely to fail due to the complexities of RAC than due to a
hardware failure.

In Case B unless your overall required uptime for the system is significantly less than
the MTBF (mean time between failures) for the aggregate of the failure rate due to all
components that cannot be repaired on the fly, RAC wins because eventually you need an
outage for preventive maintenance or an upgrade.

Now a tiny percentage of you out there may have a compute size problem where the largest
available horsepower (including descaling due to memory bus and all other hardware
contentions) in a single box is not enough. For that case RAC wins, but will probably
be more expensive and failure prone (but by a declining amount asymptotically to a
non-zero but possibly insignificant value above the failure of a single box) than a
single box if one big enough existed.

The other case where RAC has an advantage is when the size of the compute problem is
currently modest but has a substantial chance to change upward at an unpredictable rate.

Then you can add nodes to RAC. Now the more nodes you have the more problems you introduce.
But that is a strawman to the growth arguement, because no one says the nodes you add have
to be the same size as your original nodes.

And remember, until we get to 47, we don't have enough messages in a thread!

Regards,

mwf

-----Original Message-----
From: oracle-l-bounce_at_freelists.org
[mailto:oracle-l-bounce_at_freelists.org]On Behalf Of Yechiel Adar Sent: Friday, June 03, 2005 6:22 AM
Cc: oracle-l_at_freelists.org
Subject: Re: Oracle RAC cost justification?

We are going with RAC now due to business needs. We are a medium bank, ~100 branches, and we are going to implement an imaging system for checks.
All the checks are sent to the information center, and are scanned and the images are put in Oracle db.
Each morning about 3-4 people in each branch needs to go over the checks that were issued against the accounts they manage. Then they OK or reject the checks.
If the server goes down during that period all hell breaks loose. So we are going with RAC and TAF, since most of the work is reading data from the database.

Oh, the database is on a SAN, with real time copy to DR site, over fiber.

Adar Yechiel
Rechovot, Israel

Larry Elkins wrote:

>And though it strays a bit at times, but not very far, I've found this
>thread very interesting, even if it is 28 emails so far (more now I
>suppose). I am very interested in what people are seeing in the real world
>regarding RAC, and why they chose it, why they like/regret it, and
>alternatives that have been considered. So yeah, I don't know diddly about
>RAC ;-) That's why I find this thread interesting, it is very educational
>;-) I can read all the docs and white papers until I'm blue in the face but
>real world takes on it are nice to hear.
>
>Regards,
>
>Larry G. Elkins
>elkinsl_at_flash.net
>214.954.1781
>
>
>
>
>

--
http://www.freelists.org/webpage/oracle-l


--
http://www.freelists.org/webpage/oracle-l