SUMMARY: Solaris 8 Clustering / High Availability

From: Craig Scott <craig.scott_at_stc.ac.uk>
Date: Wed Jan 16 2002 - 11:56:40 EST
Thanks to all who replied 
 
John Phillips
Joe Fletcher 
Harald Husemann
Mark Hargrave
Richard Place
Eliezer Ramm
Gene Beaird
Cal Wooten
Dave Weis
Jessie Trucks
Eric Clark
Doug Otto
Jed Dobson
 
Most people suggested the use of something like Sun or Veritas's cluster
software and an external stirage device. However, while those would
ensure the databse was always(ish) available i.e. if one server fails
data on external storage can be accessed via other server giving instant
fail over. It would not fulfill the requirement to have a second
physical copy of the data that can be used for reporting and
investigation purposes, which is currently a greater operational
requirement then instant recovery.
 
A few people suggested using an Oracle standby database (one of options
already being investigated and mentioned original  positing). In this
situation the master oracle database automatically sends its completed
transactions to nominated standby databases, those database then apply
those transactions to themselves result in synchronisation with the
master database. The standby database can be placed in read only mode at
any time and then reconnected or simply disconnected to become a master
in its own right. 
 
Implementation of this option should allow two synchronised copies of
the data to exist on two physicaly seperate Sun boxes without using
shared storage. Ideally if the primary server failed we would have
little or no data loss.
 
This option would also cut the costs involved e.g. due to existing
resources would only need to buy one server, not one server and external
storage.
 
By request have included all recieved replies below.
 
Craig Scott
IT Development Officer
South Tyneside College
 
-----Original Message-----

From: Phillips, John [ <mailto:john.phillips@calanais.com>
mailto:john.phillips@calanais.com] 

Sent: 15 January 2002 10:43

To: 'Craig Scott'

Subject: RE: Solaris 8 Clustering / High Availability

 

Hi Craig,

you could do this using Sun or Veritas cluster and Oracle Parallel
server, mighty expensive solution imho for what you want.

Sun/Veritas cluster normally operates with a shared disk set which is
passed between the servers so there is only one copy of the database,
this gives high availability at lower costs (although the software
licenses are

expensive)

Oracle has a snapshot facility which could keep another database up to
date which can be used without clustering etc, you could manually bring
up an ethernet interface to take the identity of the failed server.

Finally the q&d solution would be just to copy the database files across
the wire every night..

Sounds like your manager is thinging of some of the novell products like
standby server etc, which are fine in a file and print environment...

Cheers

John

(John.Phillips@calanais.com) Unix Support, Calanais Ltd Internal Phone:
700 2643 External Phone: 0141 568 2643

-----Original Message-----

From: Fletcher, Joe [ <mailto:joe.fletcher@Metapack.com>
mailto:joe.fletcher@Metapack.com] 

Sent: 15 January 2002 10:41

To: 'Craig Scott'

Subject: RE: Solaris 8 Clustering / High Availability

 

 

You might also consider some kind of SAN solution for your disk
miroring. We run a failover rig here using shared A1000 arrays dual
hosted between servers and RSF-1 software to handle the failover. See
<outbind://3/www.high-availability.com> www.high-availability.com for
details on this.

-----Original Message-----

From: harald.husemann@Materna.DE [ <mailto:harald.husemann@Materna.DE>
mailto:harald.husemann@Materna.DE] 

Sent: 15 January 2002 11:37

To: craig.scott@stc.ac.uk

Cc: michael.schulte@Materna.DE

Subject: AW: Solaris 8 Clustering / High Availability

 

Hi Craig,

in my opinion, SUN Cluster 3.0 would be the best solution.... We have
such a system running here, including an Oracle Database. You would need
two E420 or similiar, and at least one D1000 Storage as shared storage.
Clustering without a shared storage is in my opinion impossible.... For
"real" High Availibility you would of course need two storage devices. I
would not suggest to run it without the storages, even if it's possible,
because restoring a database could take many, many time (several hours
downtime, if you have BIG tables)

best regards, hope you get it running,

Harald Husemann

System Administrator

Materna GmbH

-----Original Message-----

From: Mark Hargrave [ <mailto:hargrme@wisdom.maf.nasa.gov>
mailto:hargrme@wisdom.maf.nasa.gov] 

Sent: 15 January 2002 12:15

To: craig.scott@stc.ac.uk

Subject: Re: Solaris 8 Clustering / High Availability

 

Craig,

Checkout ORACLE RMAN! Our DBA people use this type of

backup of the database. I believe you can do full and incremental
backups of the database to one file and use this file to recover.

Thanks,

Mark

 

-----Original Message-----

From: Place, Richard [ <mailto:rplace@uslec.com>
mailto:rplace@uslec.com] 

Sent: 15 January 2002 13:42

To: 'Craig Scott'

Subject: RE: Solaris 8 Clustering / High Availability

 

Wow, seems as if you have a daunting task, with some specific
requirements to meet. The first thing that really jumps out at me is
that you are not using shared storage (storage accessible by both
servers) This seems a bad idea in that if your primary server goes down
you'll have no access to it's data from your secondary server. If cost
is an issue with regards to this implementation there are alternatives
to putting in a SAN and having an expensive array. A simple Sun A5200
would do the trick and perform as well as your internal disks. Secondly,
will someone be performing a manual failover of these systems in the
event of failure (software or hardware) or are you looking into a
packaged clustering solution? Again this depends on your budget, but
here is my recommendation for a great solution:

Connect your servers (two channels each) to an external disk array. Use
Veritas Cluster Server and Veritas Foundation Suite (cluster server has
a clustered volume manager which will handle the interaction of your two
systems to the external array) And you may want to look into Oracle
Parallel Server.

Your budget is the key here. There are cheaper ways to accomplish this
but I really would invest in shared storage for these two machines. If
you lose one, you could very possible lose a significant amount of data.

Let me know if this helps,

Rick Place 

Sr. Unix Systems Administrator

US LEC of North Carolina

-----Original Message-----

From: Eliezer Ramm [ <mailto:eramm@omnisky.com>
mailto:eramm@omnisky.com] 

Sent: 15 January 2002 14:23

To: 'Craig Scott'

Subject: Solaris 8 Clustering / High Availability

 

just MHO, but the way to do this is 2 setup 2 oracle machines connected
to an external storage device. many of these devices come w/ internal
mirroring systems that will make your life easier. on the server side
you just need identical installs of the OS and Oracle. your table space
is on the external disks. if one server fails, the other server can pick
up w/ the data in place. you can use veritas cluster for HA if needed. i
know your boss doesn't like this but i think it's the correct way to do
what you need.

there was a relevent article in sysadmin September 2001 Volume 10 Number
9 

"Creating a Cold Standby E250 " unfortunatly the article is not online.

anyway it seems that since you are willing to use cold backups what is
the problem w/ creating scripts to dump the DB and have your standby
server import them via a few scripts and scp. i don't know why you need
the gigabyte direct connection to do this.

anyway just a few thoughts.

eramm

-----Original Message-----

From: gene.g.beaird@mail.sprint.com [
<mailto:gene.g.beaird@mail.sprint.com>
mailto:gene.g.beaird@mail.sprint.com] 

Sent: 15 January 2002 14:39

To: craig.scott@stc.ac.uk

Subject: RE: sunmanagers digest, Vol 1 #972 - 18 msgs

 

Craig,

You will probably want a pair of servers, a pair of SANs or RAID arrays 

arranged in a cluster pair, and use Oracle Parallel server (Oracle 8i) 

or Oracle RAC (Oracle 9i) and Sun Cluster Server 3.0. We were going to 

configure a similar system until our budget (or lack thereof) hit us. 

You can use some of the tools that comes with the Oracle suite to make 

archives of your database and move them to either another server, or 

replicate them to the other clustered array. 

The hardware will allow the servers/arrays to talk back and forth to 

one another via fiber, and you can set up the systems to automatically 

handle fail-overs. For your testing, you can probably 

manually-fail-over a server, and work with it offline. 

Veritas also has a package called Database Edition for Oracle, but it 

is relatively new, and not certified on a lot of Sun hardware yet. HTH.

Gene Beaird

Systems Integrator V

Sprint E|Solutions

Houston, Texas

-----Original Message-----

From: Cal Wooten [ <mailto:cwooten7@attbi.com>
mailto:cwooten7@attbi.com] 

Sent: 15 January 2002 14:46

To: Craig Scott

Subject: RE: Solaris 8 Clustering / High Availability

 

Craig,

All you need to do is add a few initialization parameters and set up a
standby database. The hot database transfers redo logs across the
network and applies them. The standby database is running in essentially
recovery mode. If the primary fails then just start the standby up as
normal. It will finish its recovery and will be current as of the last
transfer. Feel free to contact me for more information. This is a very
simple solution and easy to maintain and doesn't require clustering.
Avoid parallel server like the plague (clustering). Cal Wooten

-----Original Message-----

From: Dave Weis [ <mailto:djweis@sjdjweis.com>
mailto:djweis@sjdjweis.com] 

Sent: 15 January 2002 14:06

To: Craig Scott

Subject: Re: Solaris 8 Clustering / High Availability

 

 

Hello

We do something similar with Oracle, but using standard edition instead
of enterprise edition. The main database server uses archive logging and
the logs are scp'ed over to the secondary server. The secondary server
is set up as a standby database in manual recovery mode. Oracle has a
good pdf/book on it at their site. The title is "Standby Database
Concepts and Administration". It's listed as being for version 8.1.6,
but it should be applicable to 8.1.5-8.1.7 at the least.

You can run the standby database in read only mode without breaking the
standby setup for reporting, etc. Once you activate the standby, it's no
longer a standby and you have to basically recopy the data files back
and forth, etc.

If you want to spend the money on enterprise edition, it will automate a
lot of this for you, but based on your requirements, I wouldn't use it.

Good luck

dave

-----Original Message-----

From: TRUCKS, JESSE (AIT) [ <mailto:jt9873@sbc.com>
mailto:jt9873@sbc.com] 

Sent: 15 January 2002 15:38

To: 'Craig Scott'

Subject: RE: Solaris 8 Clustering / High Availability

 

This depends on your budget. If it works for your budget, just get
Veritas Cluster Server. You can have it up and running via GUI in a few
minutes for a simple database failover. You can use it to manually fail
over or automatically. All it takes for hardware is an extra couple
Ethernet ports on each machine for a private network. Since you will not
be sharing disks, you don't even need the Veritas Volume Manager
software to make working with the disks easier.

 <http://www.veritas.com> http://www.veritas.com

 

-----Original Message-----

From: Eric Clark [ <mailto:eclark@virtual.com>
mailto:eclark@virtual.com] 

Sent: 15 January 2002 15:41

To: Craig Scott

Subject: RE: Solaris 8 Clustering / High Availability

 

Craig -

The basic problem with that type of setup is how will you replicate data
from System A to System B. There are several ways to do this. You can do
this with a cold backup in Oracle, Oracle replication, or by sending
your archive logs to the other machine, which has an inactive database
that you restore to.

Other methods are to use some type of filesystem or volume replication.
Storage hardware vendors like EMC & HDS have hardware-specific
replication software that is really good, but requires like storage
hardware on either end (EMC --> EMC). Veritas makes a product called
Veritas Volume Replicator which I also really like & it will work with
any storage hardware. VVR sends changed block info to the standby
system. The rdist command in Solaris may be another method, but it
replicates files, not data blocks & it has to be in crontab to run. This
could be painful with large data files.

If you want automatic failover, then you're looking at shared storage
(which can be as simple as a multipack) & either SunCluster, Veritas
Cluster Server or Integratus. VCS is generally considered to be the best
& they have the market share. 

I'd recommend you take a very close look at VVR. You may also want to
re-evaluate your need for a cluster (SunCluster, Veritas Cluster Server
or integratus). You shared disks can be as simple as a multipack...

A good book to check out is "Oracle Distributed Systems" by Charles Dye,
O'reilly & Associates. It has butterflies on the cover. This book
discusses distributed systems from Oracle's perspective.

Hope this was helpful - let me know if you have more questions...I work
for a consulting firm that does LOTS of this sort of thing...

Good luck!

-Eric Clark

 

-----Original Message-----

From: Doug Otto [ <mailto:doug.otto@govstor.com>
mailto:doug.otto@govstor.com] 

Sent: 15 January 2002 16:29

To: 'Craig Scott'

Subject: RE: Solaris 8 Clustering / High Availability

 

Take a look at Sun's SNDR & Instant Image products. They can do what
you're requesting. 

-----Original Message-----

From: Jed Dobson [ <mailto:jed@wgtech.com> mailto:jed@wgtech.com] 

Sent: 15 January 2002 16:52

To: Craig Scott

Subject: Re: Solaris 8 Clustering / High Availability

 

 

Use local storage, thats fine. SNDR and II can copy these data files
over the network to the other machine. Other options include NFS to
mount filesystems, dump the files to the other hosts storage then start
up.

A real nice setup would be storage attached via FC switches to both
hosts. You can have a two initiators per zone which would be perfect for
you.

T3 + FC switches would do that.

-jed
_______________________________________________
sunmanagers mailing list
sunmanagers@sunmanagers.org
http://www.sunmanagers.org/mailman/listinfo/sunmanagers
Received on Wed Jan 16 10:57:24 2002

This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:42:32 EST