Re: Data replication mechanisms?

In article <9fe1f2ad.0206061053.3ea16d73_at_posting.google.com>, laredotornado_at_zipmail.com says...
>
>Hello,
> I work for a company doing its own web hosting. Currently we are
>running an ORacle 8 cluster across 3 Sun 280R boxes. The setup is
>such that if any of the 3 boxes goes down, the others will have
>redundant data. We expect significant traffic but are noticing
>significant latency as data replication takes place.
>
>SharePlex has been recommended as a potential data replication tool.
>Can anyone suggest some alternatives or give testamonials as to why
>SharePlex is the best thing out there?
>
>Thanks in advance, Dave

Seems to me I should have a template of this response somewhere, since I need to roll it out on a regular basis.

There are two things you need to look at in this scenario - machine failover and site failover capabilities. Let's look at those in turn:

1. Machine failover can be handled in a variety of different mechanisms. Usually, it requires a cluster at the hardware level. You can then use failover at the OS level (node 2 takes up the IP address of node 1, and you manually restart the database). An alternative is to use OPS/RAC. These technologies allow you to AUTOMATICALLY fail the database over, and optionally continue queries or sessions on the second node. This leads to lower downtime than the manual restart mechanism. I notice you mention Oracle8 in your posting. OPS is the technology in Oracle8 that allows that. It's great for high availability but can be challenging to get scalability out of without additional work (note I said "challenging", not "impossible"). If 9i was a possibility, I would definitely recommend it to allow the use of RAC which is much more straightforward to get high availability AND scalability from.
2. Site failover can be handled by advanced replication, DataGuard or Logical Standby within the Oracle technology stack. Replication can be set up so that the remote site is an exact duplica of the local site, or as a master slave type relationship. DataGuard (the old standby database from Oracle8 days) allows a number of different levels at which data can be copied to the remote site, ranging from redo log shipping at the coarsest level to synchronous log vector shipping at the finest level of granularity. Logical Standby is something I haven't looked at a lot yet (it comes with 9i Release 2) but from what I understand of it it's basically a way of subsetting what is applied to the remote site.

There are other options that are available to you as well. Some (like Shareplex) come from other vendors so I can't comment on them too much other than to say that I believe that Shareplex uses its own log mining technology to read through the redo logs and is therefore completely unsupported by Oracle. Others from an Oracle persepctive might include the CDC technology in Oracle9 or LogMiner.

What you really need to do though is get someone who has a lot of understanding of HA to look at your requirements in some detail, which is obviously not an option over a newsgroup posting. FYI, I would expect someone to take at least a week to get the right level of understanding of your requirements before they come up with a solution that meets all your needs, but at least this should start you on your way. As an aside, and I don't mean to put you down here, the reason I suggest having someone else look at this is that the indications from your posting (significant latency due to data raplication) are that you don't necessarily understand HA technology well enough. In a clustered environment, there is no need for data replication to provide machine failover capability.

Let me know if you need more info.

HTH. Additions and corrections welcome.

Pete

SELECT standard_disclaimer, witty_remark FROM company_requirements; Received on Thu Jun 06 2002 - 22:26:07 CDT