Question is there any active, active cluster?

[dietcokebottle]

Hey,

Just curious is there any active active cluster solutions I kn ow microsofts is active,passive.

 

[Sleighbor]

The Microsoft solution supports active/active. I have been running it in production for over a year on my file server and SQL servers.

Kevin

 

[dietcokebottle]

but still theres downtime involved if you have to bring a server down for just say a service pack upgrade there is still user intervention I am tryingto find a solution that requires no user intervention

 

[xmsre]

In the MSCS active/active scenario, you would move the group on one node to the other node prior to applying a service pack. The only down time involved is the services going offline on the first node and coming online on the second node; the shutdown and startup time for the application. In the event of a failure, the group goes offline and comes online on the other node automagically; there is no user intervention. There is still a small amount of downtime while the services go offline and come online. This is the nature of a failover cluster with shared storage. Competative pailover clustering products suffer the same fate; it's part and parcel to the concept of failover clustering.

The other type of clustering is network load balancing. In this scenario each node runs independtly, but registers with the NLB service. NLB directs client connections to one of the nodes in the cluster. If a cluster node fails, client connections to that node are lost, NLB reconizes the node is gone, and new connections are directed to the remaining nodes. Each server hosts the complete application. There is no shared storage. Typically front end applications are load balanced, and all the front end nodes connect to the same database which is usually on a failover cluster. Application clustering by various vendors usually takes this route. Still, if you failover the backend database you will incur some downtime.

A hybrid approach adds in replication. In this approach you have NLB on the front end, and multiple backend databases where the data is replicated. This approach is harware heavy and bandwidth intensive, but it does minimize downtime. Still, there can be issues with lost transactions, conflicts, and synchroniztion.

I guess the bottom line is that you can reduce the downtime, but never really eliminate it entirely.

 

[wacobd]

In all honesty A/A vs. A/P is more of an administrator perference than anything. In our shop we do mainly A/P for simplicity and to keep the administrators rear out of the sling

I perfer to set up A/P in our database environment mainly because if something bad happens, it will happen here (and the vendors whose software we use require A/P setups).

If you want to setup an A/A cluster for your MS environment, go for it, there are no real configuration changes to installing the clustering services, just have additional resources on each node that point to the other node for failover purposes.

 

[TimUK]

Yes, there is. It's called Veritas Cluster Server. In my setup I typically have a 3-node active/active cluster. Multiple cluster volume groups containing virtual servers reside on node 1. Disk is presented over the SAN. Nodes 2 and 3 also have their own volume groups and virtual servers.

So say I take a chainsaw to node 1. The volume groups offline and online on the next available node. There is no user intervention required and it takes 90 seconds to failover. The users don't actually notice. You can cluster file/print, Exchange, SQL, Oracle etc etc etc.

http://www.veritas.com

Tim

 

[xmsre]

There is no user intervention required and it takes 90 seconds to failover. The users don't actually notice. You can cluster file/print, Exchange, SQL, Oracle etc etc etc."

Nor is there any user intervention when you use MSCS. Veritas uses data replication for distance clustering. The replication piece still has to go through a recovery process to make sure no tranactions are missed. In fact, in real world scenarios, only logs are replicated synchronously and the database would be asynchronous. Typically this means you also have to wait for log file replay on service startup. Still, it takes some time for all these activities to occur after the failure of a node. In this case, 90 seconds. EMC has similar distance clustering products, and NSI Software makes geocluster and doubletake. The difference between these and MSCS is that MSCS uses shared storage while these examples rely on data replication. It's worthy of note to mention that synchronous data replication, which you must use for the log files at a minimum when replication tranactional applications, has a severe negative impact on disk IO performance. Generally, neither data relication nor shared storage clustering solutions protect against data corruption. A good backup and disater recovery plan protects you from data corruption. MSCS allows fault tolerance for failure of a physical node. MSCS is limited by the shared SCSI bus, in that both nodes have to be in the same physical location. If the building were to be burned down, all would be lost. Distance clustering extends MSCS in that with distance clustering, the nodes can be in seperate physical locations.

 

[bsla]

We have use Matathon Technologies to Cluster some applications. This is a true active\active cluster. The applications run in lock-step on separate nodes of the cluster. If one node failes it's continues on the other. No failover needed. Very cool stuff.