Oracle DB and SAN

[Sylvain CROUET]

Hi!

I would like to know if someone has already saved an Oracle DB with a SAN architecture, without any other product like RAC or VERITAS.
The idea is: I have two servers with an Oracle DB. Server A is the main server and server B is on a backup site, in standby mode. I would like to save my Oracle DB on server A to server B, without stopping the database, and frequently enough in order not to lose more than one hour of data. Is this possible?

[Michael Steele_2]

You can you export and set up hourly ftp crons?

[Sylvain CROUET]

I do not think FTP would be the solution regarding the database consistency.
I think that a bit to bit copy through the SAN would be more efficient.

[Steven E. Protter · Accepted solution]

Shalom,

Suggestions:
1) Oracle dataguard. Does block replication from one system to another. Much less i/o and band width requirements than Serviceguard clustering.
2) Serviceguard clustering. High availability, you can take backups to other disk storage(hot) any time you want by putting the database into backup mode.
3) Oracle streams. Takes the sql being run on production database and runs it on a secondary copy database. Usually the sql is much less in size than say copying the data blocks.

SEP

[Sylvain CROUET]

Does Oracle Dataguard and Oracle streams work with an Oracle 9.x DB? Is there any distance limit with this solutions (eg: backup should not be located more than 50 kilometers from primary)?

Is it possible to have a correct backup with Oracle and SAN only, without any additional software?

[Indira Aramandla]

Hi Sylvain,

Oracle Streams was introduced since Oracle 9i release 2 (9.2). This enables the sharing of data and events in a data stream, either within a database or from one database to another

Here is in brief "when to use Streams and when to use Data Guard".

Streams are used for Information Sharing/Distribution, Multi-Directional, Data sub-setting. It allows heterogeneous platforms. Tablespace can be READ WRITE. Allows data to be shared between oracle and non-oracle datastores.
Data Guard is used for Disaster Recovery and High Availability, It is uni-directional. Provides maximum Data Protection. Automated switchover and failover. But only Homogenous platform are permitted. Tablespace should be READ ONLY. Redo Apply and SQL Apply can be done with Oracle databases only
In summary, if you require a logical standby database for data protection, use Oracle Data Guard. Oracle Data Guard is designed specifically for the creation and management of a logical standby database. It automates many of the common tasks, reducing the need for custom management scripts. On the other hand, Streams provides much finer granularity and control over what is replicated and how it is replicated. Streams supports bi-directional replication, transformations and custom apply functions. It also gives users complete control over the routing of change records from the primary database to a standby database. This allows users to build hub and spoke network configurations that can support hundreds of replica databases.

Refer to Note:300223.1 in Oracle metalink for "Comparative Study between Oracle Streams and Oracle Data Guard"


Indira A

[Frank de Vries]

Hi,

We use business to bussiness copy (that is what it is called ) but it is effectively
a bit to bit copy of partititions attached
to the lun's.

The software comes from HP and runs on XP256 or XP512 (Basically an HITACHI technology)

The only snag is that we do not use a standby mode. We do not trust the write behind buffers, and stop the db for a tick and bring it up completely mirrored.

[Sylvain CROUET]

Hi Franck!

Can you be more precise? What is the name of the software solution? How is your DB configured?

[Joshua Small_2]

The do it yourself solution involves a regular backup (no reason you can't do this online) and having Oracle write Archive logs to a remote server.

These can be written locally and rsync'ed to another box, and despite not involving a big name product, is quite a reliable procedure.

[Frank de Vries]

Hi Sylvain

We use the HORCM raid manager for Hpux
platforms , they work with the XP256 and XP512 systems.

Please find attached the conf file with
loads of technical details. That will give
you more inside into how the Raid manager
uses LVM linked to LUN to setup
business-to-business copy at disk level.

If you need more info let me know.

[Frank de Vries]

I forgot to mention that with the command
set you create a pair,
paircreate source lun target lun
Once the pair is setup you can activate
for the type of raid-mirroring

I will give the output of pairdisplay that
will explain a lot:
[root@orasrv2:]/HORCM/usr/bin<>>> pairdisplay -h
Model : Raid-Manager-XP/HP-UX
Ver&Rev: 01.08.00
Usage : pairdisplay [options] for HORC
-h Help/Usage
-z Set to the interactive mode
-zx Set to the interactive mode and HORCM monitoring
-q Quit(Return to main())
-g Specify the group_name
-d Specify the pair_volume_name
-d[g] [mun#] Specify the raw_device without '-g' option
-d[g] [mun#] Specify the LDEV# in the RAID without '-g' option
-c Specify the pair_check
-l Specify the local only
-m Specify the display_mode(cas/all) for cascading configuration
-f[x] Specify display of the LDEV#(hex)
-f[c] Specify display of the COPY rate
-f[d] Specify display of the Device file name
-CLI Specify display of CLI format
-FCA Specify the force operation for cascading CA_VOL
-FBC [mun#] Specify the force operation for cascading BC_VOL

Anyway, once your raid-pair is activated,
it is possible to failover (by the command horctakeover) to the target pair (which effectively is seen from a different node.

This works very well, but like I said, it is
still safer to shut the DB's, as you can never be certain the buffers will be written to datafiles during the takeover is finished.

We run Mc/Serviguard on top so we have the flexibility to not only swith nodes but also between the 2 XP systems. A fully fledges crossed-linked san failover system.

Basically plain-crash proof !!!
(Our Nodes and XP are in different buildings,
linked up with fibre)

Anyway I think you get the picture :))

[Sylvain CROUET]

What is the distance between you buildings? Would your solution still works over 50 km?

[Frank de Vries]

That is possible:
Depends on your fibre is multi or single mode and the thickness. See fragment below
I took from the Cisco manual for Certification:

What is the maximum distance fiber optic transmitters can operate at?
It depends on which LuxLink(tm) model you purchase. Normal transmission distances can vary from a fraction of a mile to 40 miles (60 Kilometers) or more. The maximum transmission distance depends on output optical power of the transmitter, the optical wavelength utilized, the quality of the fiber optic cable and the sensitivity of the optical receiver. In general single-mode based systems operate over longer distances than multimode systems. The approximate transmission distances for LuxLink(tm) systems are indicated in the table below.

-No. Wavelength Fiber Type Connector Transmission Distance covered**

-1 850 nm multimode ST up to 2 miles (3 Km)
-3 1310 nm multimode ST up to 6 miles (10 Km)
-7 1310 nm single-mode FCPC up to 20 miles (30 Km)
-8* 1310 nm single-mode ST up to 20 miles (30 Km)
-9 1550 nm single-mode FCPC up to 40 miles (60 Km)


regards,

[Sylvain CROUET]

If the distance between my two sites is over 100km, what kind of solution can I use?

[Frank de Vries]

I see what you are getting at.

Well then you need to inmplement an intermediant groundstation, that amplifies the signal. Equipment like this can be purchased from Nortel networks for instance.

Or alternatively, you can rent a public or private line from a telecom provider and they will do it for your.

We hire a T1 line from French telcom that connects our Brussels servers with our Stockholm server, were the standby server is , and no one realises that. It looks as if it stand in the room next door. Just a matter of bandwith and budget !!
(Technology is not the restraining factor ,
believe me)

[Sylvain CROUET]

You are not faced with delays due to the distance? You still have a real time bit to bit copy of the database?

[Frank de Vries]

Well the functional answer to your question
is no, no delays at all, as to user it is transparent.

However, for the nitty gritty technician , naturally there is an propogation delay due to distance and ground equipment and corresponding dataload. But this
propogation delay is measured in a dozen or
so milliseconds, so that is neglectable.

To give an example. A link via Satelite,
for large video images , thus up 37thousand km up, and again 37K downlink, takes 0.75 second to do the roundtrip.
So what would the difference between 10km and
100 km for a data packet, it is transparent for the user.

More important is that the bandwidth is not
saturated. You can have longer delays on
a local lan.

Also , Of course we have a reduntant line via an alternative path, with automatic switchover if timeouts are not respected.

[Sylvain CROUET]

So, if I summarize your configuration: you have two servers with a database, a primary server and a backup server. Each server is connected to a SAN built with XP arrays. Bit to bit replication is made in real time by HP StorageWork Business Copy XP between the XP arrays. There is no impact on production.
Am I right?