Re: with all this talk about RAC...

  • From: Matthew Zito <mzito@xxxxxxxxxxx>
  • To: oracle-l@xxxxxxxxxxxxx
  • Date: Tue, 25 May 2004 11:55:00 -0400

Bill,

My company has made a business out of both building software around 
clustering technologies such as RAC, as well as offering professional 
services related to RAC.  As such, we've spent a lot of time fixing 
failed RAC implementations, building a set of optimizations and 
best-practices around RAC, and helping organizations figure out when 
clustering is the right solution.

As other people have said, the pre-eminent reason for RAC that has 
proven to be the most dependable is the high-availability features.  
Active/active clustering is a huge advantage, not just in 
"mission-critical" systems, but in any organization that may not want 
to be under pressure to repair a failed system in any particular 
timeframe.

The other reason that holds true more often than one might think is the 
scalability advantages.  Certainly, it would be absurd to expect to 
replace a 40-processor Sun server with 20 2-processor servers, but it 
is not unreasonable to replace an 8-processor sun server with 2 
4-processor servers, or even 3-4 2-processor servers.  As technologies 
such as InfiniBand and RDMA-over-Ethernet mature, it is going to be 
easier and easier to scale these clusters larger and larger.

The last advantage is the cost - the cost of the RAC licenses is often 
touted as eating up the hardware savings (plus the added complexity of 
administering RAC), but hard negotiation can bring that cost down to 
minimal levels.  A second point is that the cost difference of the RAC 
licenses can often be absorbed through the processor savings you can 
get by moving from last-generation processors to current ones.  The 
administration complexity is a valid concern, and one of the reason 
that there are a number of companies (mine included) that have software 
to reduce that complexity.

That's actually the reason we see so many failed RAC implementations - 
because clustering is fundamentally difficult, requires a much greater 
level of OS, storage, and hardware involvement than a normal database 
instance, and Oracle's documentation is sketchy and incomplete.

RAC is not a silver bullet - but what we've found is that more 
organizations than not could be using RAC to reduce cost and improve 
uptime.  It's certainly a technology worth looking at.

One last tech note - I can not recommend RAC on Sun - not only is the 
hardware more expensive, removing a lot of the cost savings, but the 
additional software required can be expensive.  Plus, there's a great 
deal more complexity - you need Oracle, Solaris, Sun cluster, and 
veritas, and you have to use raw devices (unless the sun clustered 
filesystem has been qualified - I haven't checked in a while).  You end 
up with a very complex implementation, multiple vendors to deal with, 
and limitations on configuration.  On Linux, you end up working with 
only the single vendor for the end-to-end cluster stack.

Thanks,
Matt

--
Matthew Zito
GridApp Systems
Email: mzito@xxxxxxxxxxx
Cell: 646-220-3551
Phone: 212-358-8211 x 359
http://www.gridapp.com


On May 25, 2004, at 8:15 AM, bill thater wrote:

> i have a few serious questions.
>
> as the last part of a major upgrade of hardware/OS/software in our 
> production hosted environment, we're getting to upgrade the database 
> cluster.  now we have a couple ways to go here, linux cluster or sun 
> cluster, but i can sit here and make a good case both for and against 
> using RAC for our production databases.
>
> so what is the collective wisdom on the deciding factors to implement 
> RAC?
>

----------------------------------------------------------------
Please see the official ORACLE-L FAQ: http://www.orafaq.com
----------------------------------------------------------------
To unsubscribe send email to:  oracle-l-request@xxxxxxxxxxxxx
put 'unsubscribe' in the subject line.
--
Archives are at //www.freelists.org/archives/oracle-l/
FAQ is at //www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------

Other related posts: