Ask the Expert

Replication: Sun vs. Veritas

We have eight database servers, two pairs with normal Sun Cluster, and the other two are inter-domain clustered also using Sun Cluster. During our next phase, we are going to have more databases. The servers are currently running different OS and VxVM versions.

I have two proposals from vendors. First is to use host-based replication (VVR with GCO) with all new servers and storage. But we need to upgrade all the OS and VxVM to same level in order to get the best VVR and GCO features, and also have to install Veritas Cluster.

Second is to use Sun's TrueCopy, move SE9900 for development to our DR site and buy new servers and replace DEV storage with something lower end. This way, we don't need to worry about the versions. I would like to implement the second option. The most important thing is the manageability and ease of maintenance. Which option do you suggest (and why)? Which one is better in term of manageability and operations?

    Requires Free Membership to View

    Login

I am a fan of using good (but not the extreme high-end) hardware with software-based replication. I have seen too many expensive, high-end storage arrays with hidden or poorly documented single points of failure, such as backplanes. Also, if the replication function was added on to the array's firmware, and is not native to the hardware; it can be less feature-rich than a product that was designed specifically for replication.

Once either kind of replication is set up and functioning, it should not require very much regular maintenance, so I don't think that should enter the decision process. (Note: I have not worked day-to-day with TrueCopy, and that's really the only way to know the degree of regular maintenance that a product requires.)

If the products you are looking to use require identical OS and VxVM revision levels, you should certainly meet those requirements. Even if it is not absolutely required, it is an excellent practice to run the same versions of the OS and of all system and otherwise critical applications on clustered systems. You don't want some subtle difference or incompatibility to bite you. Besides, how likely is it that your vendor has tested every combination of mixed versions that you might need?

You should also consider how much data will be lost in the event of a disaster. If replication is being done synchronously, you should only lose transactions that were in process when the failure occurred. If the replication is asynchronous or semi-synchronous, then more transactions can be lost.

One final consideration, and one that is often overlooked, is to think about what happens after the disaster. Can you use the remote systems and storage as the new primary? Are there limitations? What if you want to switch back to the old configuration? I know that VVR does that very well.

This was first published in May 2005

Join the conversationComment

Share
Comments

    Results

    Contribute to the conversation

    All fields are required. Comments will appear at the bottom of the article.
    Back to top