There can be three main reasons for using clusters (I think):
- High availability. But how often does a Unix or Windows2000 server crash these days? Now, if you have servers that crash before you cluster them, you're exactly going to help things by clustering several servers that are crash prone, are you? I mean: 0.9 x 0.9 is not 2.1 :-)). So: If you have servers that crash now and then, you should probably get that fixed first - otherwise you'll be very disappointed by clustering them. It's not exactly something that is widely published by IBM or other vendors, but when you move from a standalone AIX server to a two-node HA/CMP cluster you also move from 99.9% availability to 98% availability. This is known by the good techies everywhere inside vendor organisations. There are several reasons for this, which I'll get back to in a second...
- Cpu scaling. When Digial made their original VAX clusters it was partly because you could only put one cpu in each box. That argument doesn't really count in the Unix world these days, but it might have come back with the relatively few cpus supported on Windows this and Windows that.
- Application partitioning: Let the OLTP users run against box A and the batch users against box B.
Now, when people want to run clusters (for whatever reason) and run into problems, it's usually due to:
- The added complexity. You add three layers (more or less) to the technology stack. Does that make anything more stable ;-) ? Just think about shared disks, which works fantastic, by the way, in VMS, but not really anywhere else (yeah, I use every chance to make fun of Unix, although I know VMS is dead).
- The additional management burden. As you guys have found out, it's pretty hard to understand, setup, administer, manage and - especially - troubleshoot.
- The added people skills required. You don't just run clusters and/or OPS/RAC as an average DBA. That's why we at Miracle really hope the sales of these technologies will take off, since we have some of the few guys here in Denmark who know about it :-))).
Here's something else I've been wondering about: If the rather smart folks at the various Unix vendors (they're hardly any stupider than us on this list, do you think?) cannot get this stuff to work after having tried for many years - why do people then beleive that Microsoft can get it to work just like that? That's the beauty of people: They keep beleiving in the next technical fix.
Most OPS-sites I have known about thoughout the ages have gone to standby solutions (called DataGuard in Oracle9i, I think) and dropped the OPS/cluster stuff.
Mogens
Gene Sais wrote:
I also heard of horror stories regarding Sun Clusters. I worked w/ HP MC Service guard, good product. Now working w/ IBM HACMP, also good product, although more complicated to set up (but then again I am not a IBM'er). IBM tends to do everything their way ;).
In the future when I upgrade to 9i, I will use Oracle's Data Guard or maybe look at a 3rd party product such as shareplex (good reviews, but pricey).
Gene[EMAIL PROTECTED] 01/17/02 01:05PM >>>Thanks for your help everyone. Very useful advice, although your scaring me
of Sun Clusters.
At the minute, Parallel server looks the best, with a standby database
remotely for disaster.
Does anyone know what the HP solution is like (MC Service Guard)? I think
some one on this list gave it a good review in the past .
Thanks,
Jim
-----Original Message-----
Sent: 17 January 2002 17:12
To: Multiple recipients of list ORACLE-L
IBM HACMP works well.
Ooops. guess that means you'll have to change some things. ;-)
Seriously, we *did* get the Sun "clustering" working, but it
required some serious feet-to-fire holding and gyrations.
-----Original Message-----
Sent: Thursday, January 17, 2002 11:54 AM
To: Multiple recipients of list ORACLE-L
Thanks for the advice everyone.
So what do you recommend on a Sun cluster/machines for failover oth er than
OPS?
Quest Shareplex?
Standby database?
Any others?
Thanks,
Jim
-----Original Message-----
Sent: 17 January 2002 16:22
To: Multiple recipients of list ORACLE-L
I concur with BB.........yea, I ran Sun "cluster" at <deleted> and
it broke ALOT.
Kept me and two full time Sun Engineers (they got paid ALOT more)
in consulting dollars, but i made a mental note not to use
it in "my business".
Caveat: this was 1.5 years ago. Things change.
Mit Gluck, mein freund.......
- Ross "mit schuss" Mohan
-----Original Message-----
Jim:
Sorry, you're not gonna like this answer. HA is a Sun product, not an
Oracle product. Under Sun's High Availability, you can configure several
modules like Sybase and Oracle. (The Oracle product is Sun Cluster HA-DBMS
for Oracle.) It does require what I believe Sun calls a cluster but (IMHO)
is a bastardization of t he term. It truly is failover, not cluster.
We've had lots of problems with it. It's caused us lots of grief, and only
in a few instances gained us anything. It is NOT OPS, as the database does
not run in parallel, but only on 1 box at a time. (Everything is double
cabled, and so the drives are re-mounted on the 2nd box if a failover
occurs.) Your users still get disconnected. You'd probably lose less data
than with a standby (since you pick up with the same drives mounted on the
other box), but it depends on how you have the standby implemented.
There's no additional cost from Oracle to run this crap, but you'll be
paying Sun great sums of money. The Sun web site has more info on HA.
Let me know if you need more info.
Good luck!
Barb----------
From: James McCann[SMTP:[EMAIL PROTECTED]]
Reply To: [EMAIL PROTECTED]
Sent: Thursday, January 17, 2002 5:40 AM
To: Multiple recipients of list ORACLE-L
Subject: Standby Instance questions and HA
Hi,
I was reading in the book "Oracle 24/7 Tips and Techniques" about
Standby
Instances.
Note, this is not a standby database.
From the book it seams to work in the following way...
There is only one database.
The database files exist on a shared disk pack. One machine is the primary
instance, and if this instance dies, a new instance is started on the
second
machine using the datafiles on the shared disk.
The problem is that I can't find anything in the Oracle docs about this,
or
on Meta Link.
< br>I also want to know if this method of HA requires a clustered environment
(I
think it does, but just want to be sure)?
Also, does it come with an Enterprise Edition license?
Or is it something which each hardware vendor implements in their own way,
at extra cost?
We have a requirement for a fail over method on Sun Solaris.
We do not want to loose any committed data (i.e. a standby database could
loose some), and want the fail over to be as automatic as possible.
We don't want the expense of Parallel Server (Anyone know how expensive it
is these days?).
The disk pack is RAID, and we may also have a standby database off site.
Has anyone any recommendations?
Thanks,
Jim
--
Please see the official ORACLE-L FAQ: http://www.orafaq.com
--
Author: James McCann
INET: [EMAIL PROTECTED]
Fat City Network Services -- (858) 538-5051 FAX: (858) 538-5051
San Diego, California -- Public Internet access / Mailing Lists
--------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from). You may
also send the HELP command for other information (like subscribing).--
Please see the official ORACLE-L FAQ: http://www.orafaq.com
--
Author: Baker, Barbara
INET: [EMAIL PROTECTED]
Fat City Network Services -- (858) 538-5051 FAX: (858) 538-5051
San Diego, California -- Public Internet access / Mailing Lists
--------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from). You may
also send the HELP command for other information (like subscribing).
--
Please see the official ORACLE-L FAQ: http://www.orafaq.com
--
Author: Mohan, Ross
INET: [EMAIL PROTECTED]
Fat City Network Services -- (858) 538-5051 FAX: (858) 538-5051
San Diego, California -- Public Internet access / Mailing Lists
--------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from). You may
also send the HELP command for other information (like subscribing).
--
Please see the official ORACLE-L FAQ: http://www.orafaq.com
--
Author: James McCannI was reading in the book "OracINET: [EMAIL PROTECTED]
Fat City Network Services -- (858) 538-5051 FAX: (858) 538-5051
San Diego, California -- Public Internet access / Mailing Lists
--------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from). You may
also send the HELP command for other information (like subscribing).
--
Please see the official ORACLE-L FAQ: http://www.orafaq.com
--
Author: Mohan, Ross
INET: [EMAIL PROTECTED]
Fat City Network Services -- (858) 538-5051 FAX: (858) 538-5051
San Diego, California -- Public Internet access / Mailing Lists
--------------------------------------------------------------------
To REMOVE yourself from this mailing list, send an E-Mail message
to: [EMAIL PROTECTED] (note EXACT spelling of 'ListGuru') and in
the message BODY, include a line containing: UNSUB ORACLE-L
(or the name of mailing list you want to be removed from). You may
also send the HELP command for other information (like subscribing).
