RE: intel clusters in a box

  • From: "Marquez, Chris" <CMarquez@xxxxxxxx>
  • To: <paul.baumgartel@xxxxxxxxx>, <oracle-l@xxxxxxxxxxxxx>
  • Date: Sat, 22 Jan 2005 17:39:12 -0500

Paul,

-First
>> insistent that all data replication between sites=20
>> take the form of synchronous SAN-to-SAN replication,
>> in which writes at the primary site are not confirmed
>> to the application until the data has traveled over=20
>> the fiber and been written at the remote site as well.

You will either have one of the best networks in the world
or this mandate will last about a day or less of production use.

A customer recently wanted the same thing.  We met with EMC,=20
CISCO, HITACHI and others.  Basically what I came away with is=20
if you want to mirror your SANs you have to have lots of money=20
for the best network AND STILL, there are distance and other=20
technical and logical restrictions of what you can do.

And even when you have ASYNCHRONOUS replications you have a "queue".
Any form a Replication and Queues have management hassles.

Not saying it isn't possible, but didn't find a good example in our =
meetings.


-Second


>> RAC nodes at the primary site, and one at the secondary,=20
>> all part of the same cluster.  Can this configuration,=20
>> can all nodes behave as though they are connected=20
>> to the *same* physical array?

Simply no.  This config would introduce so much latency at both the=20
disk array level and network (RAC interconnect, aka "Cache Fusion")
the performance would be so bad that it would only be something to test=20
and never to use.

My customer told us they had heard rumors of a WAN based interconnect=20
protocol for RAC...and I guess the whole "Grid" idea (will) means =
geographical=20
distant RAC nodes, but I'm telling you this is the future...its not =
today. =20
Personally having used OPS/RAC for years...it is a technology designed=20
for two (or more) server side-by-side!

my $.02


Chris Marquez
Oracle DBA
HEYMONitor(tm) - heymonitor.com
"Oracle Monitoring & Alerting Solution"



-----Original Message-----
From:   oracle-l-bounce@xxxxxxxxxxxxx on behalf of Paul Baumgartel
Sent:   Fri 1/21/2005 3:43 PM
To:     oracle-l@xxxxxxxxxxxxx
Cc:=09
Subject:        Re: intel clusters in a box
I'm also about to start putting together a cluster for RAC on
Oracle10g.  What are the Amazons and eBays of the world using for
hardware and OS?

And if I may slip in a related question:  the Tech. Services directory
of our parent company is planning two new data centers, primary and
secondary.  There will be identical SANs at each (probably HP EVA
arrays), connected via high-speed fiber.  He is insistent that all
data replication between sites take the form of synchronous SAN-to-SAN
replication, in which writes at the primary site are not confirmed to
the application until the data has traveled over the fiber and been
written at the remote site as well.  My question is this:  Can RAC
support such a model, and how?  Let's say two RAC nodes at the primary
site, and one at the secondary, all part of the same cluster.  Can
this configuration, can all nodes behave as though they are connected
to the *same* physical array?

Thanks in advance.

On Tue, 18 Jan 2005 11:14:36 -0500, Ray Stell <stellr@xxxxxxxxxx> wrote:
>=20
> I've been building my own intel/linux clusters for testing.  I'm
> wondering if you have tested any commerical offerings for the whole 9
> yards; processers, FC, and storage.
>=20
> http://www.straightdope.com/classics/a2_252.html
>=20
> "More commonly, however, you hear that the expression comes from the
> capacity of ready-mix concrete trucks. Concrete trucks supposedly
> contain nine cubic yards when fully loaded."
> =
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
> Ray Stell       stellr@xxxxxx        (540) 231-4109        28^D
> --
> //www.freelists.org/webpage/oracle-l
>
--
//www.freelists.org/webpage/oracle-l



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

Other related posts: