You have 4 options from which you can select, or even combine all 4 in
case of havy paranoid.
- Parallel Server (now it is named Oracle Real Application Cluster
(RAC)). This gives very fast recovery time in case if one node fails.
Another node just get connections to itself. The select statements from
died node can be even continued at second node. User sessions will be
reconnected from failed instance to alive one. You have to run oracle
instance on all nodes (calculate licensing fees!).
- Failower. Looks like RAC. But secondary instance is not up. There is
special daemon on second node which monitors first node. In case of
failure, daemon starts oracle instance and listener to accept new
connections. You need shared disk array between 2 nodes to keep DB files.
- Standby. You do not have to have a shared disk array. Even better 2
boxes will be separate at all. Second database will be in continuos
restore mode. All archive logs from DB1 will be applied to DB2. Many
tasty options exist in O9i on this matter, as standby can get changes
after every transaction(!). As for 8i, your standby will be late on
1 redo log file.
- Replication. Very flexible. You can deside on table level what to
replicate. Standard R - data propagation goes only one direction, from
updateable table to readonly snapshot. Advanced replication - changes
propagation can go both ways. + data collisions can be resolved
automatically by standard or custom procedures.
i1 and i2 is suitable for failover situations. Nodes should be close to
each another. Network channel should be very fast for i1.
i3 can be usefull as for failover as for disaster scenario. Network
should be thik.
i4 is suitable for disaster.
Mark
PS Dig in technet or metalink on keywords "availability, failover, disaster"
James wrote:
> We want to implement Oralce ERP in our company. Generally, there will
> be only one database server for the system. But my bose asked me can
> we protect the application by adding one more database server? Anybody
> has experience on it?
>
> We have server cluster on hardware level already.
>
> Thanks,
> James
>
- WAS CANCELLED BY =======:
From: Mark Malakanov <markmal_no_spam_at_home.com>
Control: cancel <3C54453C.10808_at_home.com>
Subject: cmsg cancel <3C54453C.10808_at_home.com>
Date: Mon, 28 Jan 2002 01:43:51 GMT
Message-ID: <cancel.3C54453C.10808_at_home.com>
X-No-Archive: yes
Newsgroups: microsoft.test,alt.flame.niggers,comp.databases.oracle.server
NNTP-Posting-Host: w088.z064003087.lax-ca.dsl.cnc.net 64.3.87.88
Lines: 1
Path: news.uni-stuttgart.de!dns.phoenix-ag.de!newsfeed01.sul.t-online.de!newsfeed00.sul.t-online.de!t-online.de!newsfeed.gamma.ru!Gamma.RU!news.stealth.net!msrtrans1!msrnewsc1!cppssbbsa01.microsoft.com!tkmsftngp01!tkmsftngp04!u&n&a&c&anceller
Xref: news.uni-stuttgart.de control:40720983
This message was cancelled from within The Unacanceller's glorious new software, Lotus 1-2-3 For Rogue Cancellers.
Received on Sun Jan 27 2002 - 12:19:14 CST