Database Redundency and Failover

  • I'm in the process of putting together a "high availabilty" / "Failover" type scenario for my DB, Application and Web Server. I wanted to get some input from everyone. Of course money is always factor but I'm want to do a cost analysis on this with a few different solutions. I was wondering about pros and cons of these different solutions.

    I'm not familiar with the costs associated with Network Load Balancing vs Clustering and or Active/Active Failover Servers. I've been researching it a little but I'm seriously considering this for a Q4 project.

    I'd like to at least:

    1- Have the abilty to bring down (in a planned execution) our DB server at our current facilty and failover to another server (locally). This way I can do maintance (software/hardware upgrades or if something fails replace it)

    What is the lag time to bring a failover server online? What are some of the pros and cons of this?

    What in a Disaster Recovery scenario would it take to have a hosted Failover server located in another state? Ideally I'd like to have an offsite location just in case that acts similar to the local failover server.

    2. Ideally I'd like to have a network load balanced server/Cluster that would service requests if the other server fails.

    I'd appreciate any other links and or opinions anyone has. TIA.

  • we have a few clusters and a DR site in another state

    for a cluster it takes around 10 seconds to failover to the other node. downside is taht depending on the application, it may need a restart

    for DR all the critical SQL data is on an EMC SAN. we use their SRDF tech to ship the data on a disk level to another EMC SAN at another office where we have standby servers. if need be we would mount the databases there. then repoint the applications

    for read only we use replication to a server at another office and point the apps there for a R/O failover

  • We use clustering and/or mirroring for SQL 2005 depending on the application. I prefer mirroring since you have 2 copies of the data on separate disks. You have additional cost for the extra disks but if your SAN dies and you have a cluster there isn't much you can do. I believe the licensing costs for a SQL 2005 mirror are the same as a for an Active/Passive cluster.

    One issue we have run into with clustering is long failover times due to many LUNs attached to the cluster. Failover times are around 1 min. but 3rd party apps timeout after connections are dropped for 30 sec.

    Also, SQL 2008 has the ability to recovery torn pages on the Principle server by grabbing them from the Mirror server.

Viewing 3 posts - 1 through 2 (of 2 total)

You must be logged in to reply to this topic. Login to reply