Replication Architecture/HA Question

Question:

Is it possible to use an automated standby database (ASD) in a multi-master replication environment? If not, what is a good way to build intra-node failover into a large multi-master environment without using OPS or adding failover databases into the replication group?

Background:

I'm in the process of architecting a global website using Oracle 8iR2 and Advanced Replication on Linux. So far, AR fits the application requirements perfectly. The topology of the site looks something like:

• primary node in New York
• secondary node in US midwest
• tertiary nodes in Europe and South America

These are high bandwidth connections, with Foundry hardware providing global load balancing. Each of the four nodes is doing multimaster replication to the others. Each node is on UPS, redundant network connections, and utilizes SMP systems with RAID 0+1, etc.

By nature of the application, async. replication works well (global load balancing logic allows any site to fall as much as 30 minutes behind the others without affecting user experience). However, only the New York node ties to order fulfillment. The other nodes (especially tertiary) can fail without critical impact on the whole site.

To build failover into the US nodes, there seem to be two options:

1. Add two additional nodes into the replication group, build application logic to fall back to these systems if the primaries fail. The downside is the extra overhead of two additional systems doing replication -- not to mention the added complexity of conflict resolution.
2. Use ASD (log replay) to keep a warm failover system in place. Manually cut the system over in case of failure. How would this work with AR? Is it possible?

(Note that OPS isn't feasible; Linux is the OS of choice for this environment and a shared disk subsystem isn't really available.)

	Thanks in advance for any insights,
	K.Sridlhan

Sent via Deja.com http://www.deja.com/
Before you buy. Received on Sun Aug 27 2000 - 17:26:00 CDT