Re: RAC and DataWarehouse

From: JEDIDIAH <jedi_at_nomad.mishnet>
Date: Thu, 22 Jun 2006 12:23:51 -0500
Message-ID: <7a0rm3-nah.ln1@nomad.mishnet>

On 2006-06-21, DA Morgan <damorgan_at_psoug.org> wrote:
> Chuck wrote:

>> DA Morgan wrote:
>>> JEDIDIAH wrote:
>>>> On 2006-06-20, Chuck <skilover_nospam_at_bluebottle.com> wrote:
>>>>> DA Morgan wrote:
>>>>>> Here are some of the more obvious advantages.
>>>>>> 4. Incremental scalability
>>>>> Is the scalability any closer to linear than it was with OPS in 8i or is
>>>>> there still a fairly low point of diminishing returns?
>>>>     RAC is MUCH better.
>>>>
>>>>     In many cases, RAC scales better than NUMA.
>>>>
>>>>     Oracle re-engineered the part of OPS that sucked so badly.
>>> Oracle's published RAC scaling number is 84%. That is roughly what
>>> I have seen in my lab too: Usually a bit better.
>> 
>> So if scalability were your main concern, what benefit is there to RAC
>> over say, adding more CPU's to the server?
>

> Let me give you H/P's own numbers published, IIRC, by Vic Andrade
> (vic.andrade_at_hp.com).

>
>===============================================

> Add 2 850 MHz procs and 2GB RAM quoted at $150K (add to 10 CPUs)

>

> Add dual 3.0 GHz server w/4GB RAM quoted at $24K (add a 2 CPU node)

>===============================================

	...except that's not the end of the story.

	Now that you are clustered you will need some sort of shared

storage solution. You will probably want something comparable to the sort of SCSI U320 solution you might have used for a single instance Oracle database. This shared storage solution will also require additional (rather expensive) HBAs be added to each node as well as suitable redundant SAN switches.

        You will then have n systems that have to be managed and maintained and cooperate across your cluster interconnect and your shared storage system.

        You will also still not get away from the need for further redundancy. Either the highly expensive SAN storage solution will need to be redundant or you will need a standby solution.

        Your application will also need to be tolerant of single node failures or altered to become so.

>

> The price difference more than makes up for the cost of the additional
> RAC license to Oracle.

>

> Obviously the price of the new node, with current pricing would be
> less than $24K. I think you will find the pricing adding 2 CPUs to
> an existing 8+ CPU box is still very very high.

        Putting another system board into a system that operates in that manner is remarkably more transparent than adding capacity to a tightly coupled cluster like RAC. It's not exactly a Beowulf.

        RAC will require better sysadmins than NUMA will.

        If your admins don't already grok SAN and clustering, that will be an extra hurdle that RAC brings to the table.

        Linux in general tends to encourage management to cut corners where they really have no business doing so. This "Linux is cheaper" mantra really is quite problematic.

        Got a local OUG? Go there and seek out people doing RAC deployments and get their input.

-- 
	vi isn't easy to use.				 |||
							/ | \
	vi is easy to REPLACE.

 Posted Via Usenet.com Premium Usenet Newsgroup Services
----------------------------------------------------------
    ** SPEED ** RETENTION ** COMPLETION ** ANONYMITY **
----------------------------------------------------------        
                http://www.usenet.com

Received on Thu Jun 22 2006 - 12:23:51 CDT