Exchange 2010 Stretch Cluster – 1
Posted by Brajesh Panda on March 10, 2010
Here is my 1st
Architecture Diagram for Exchange 2010 Stretched Cluster. Expect my step by step guide to build this environment. I will keep improving my architecture… you guys too buzz me if there is a place for improvement ;-).
For this scenario I am considering single stretched DAG & active mailboxes only in Primary site. If we keep active mailboxes in both sites with single stretched DAG, in case of WAN failure, mailboxes in secondary site will go down. If we want to put active mailboxes in other sites too, take a look at my next article.
As this architecture contains 3 nodes (Odd number), we don’t need a File Share Witness Share folder for Quorum. We can just use Node Set Majority quorum. In this architecture I am stretching the DAG between both sites, so in case one node goes down (let suppose from primary site) there are still 2 nodes online, which helps to maintain the Cluster Quorum. Let suppose two nodes goes down, now you have just one node online. Hence DAG cluster will go down. But you can manually run switch over process & activate the secondary site DAG member using Alternate File Share witness. So in this case you are changing your quorum model to FSW, but you may like came back after other two nodes came back online. One thing to note, even if we are using Node Set Majority we can mention file share witness & alternate witness – only cluster will not be using them actively. In above scenario now we have a active server with an alternate file share witness in secondary site. What happen when Primary site servers come back online before WAN link? There are two nodes in primary site, which can provide quorum & try to mount databases. Resulting Split Brain. To avoid this there is another mechanism called Data Center Activation. If this configuration is on, before they mount the databases, they will check the status of DAG from every node using Data Center Activation Protocol & then decision will be taken if Active Manager should mount the database.
Other key thing is client access name space designing. If both sites are hosted in two different AD sites, we will be having Two different CAS Array (RPCClientServerAccess) FQDN. And if in case both CAS servers or load balancer failed in Primary site, we have to manually change the DNS IP Address of Primary Site CAS Array to secondary site CAS Array IP.
Well I just love Geo/Stretched Cluster….