Clustering in author env

Question

In my company we are using 2 publish server for internet and 2 publish server in intranet and one author server.
We have around 10 Intranet sites and 10 internet sites. We are planning to migrate 80 more sites in internet and 100 in intranet.
Content size is approx 20GB in internet and intranet it will increase it to 3 times after migration.
No of authors will be 500+ currently it is 100.
I am thinking of upgrading CQ publish to 64 bit OS and use SAN disk increase memory and cpu .Also business wants to make author env resilient so i am planning to use shared data store clustering in author env and multiple publish server ( no clustering  )
My question is will i be able to use replication and reverse replication in this case ? Please let me know your view on this approch.Do you suggest to use shared nothing clustering in author env?
high level architectural of my solution is ( except author cluster with shared data store)    
http://dev.day.com/content/docs/en/crx/current/administering/cluster/_jcr_content/par/imag e_0.img.png/1350922018297.png
Regards,
Manish

Answer 1

Whether to use shared data store or shared nothing is really dependent on why you are clustering, and whether or not your shared data storage has its own built redunecy or not.
If you are implementing the cluster to ensure high availability and diaster recovery then you need decide whether or not the shared data store is a single point of failure. Usually with shared data store the shared data store is mounted to either something like a SAN or some other remote mounted storage system. If this storage system is itself highly available then you can use the shared data store to reduce total disk usage and still have a highly available environment.
If on the other hand the data storage mechanism isn't highly available then you would want to consider the shared nothing cluster in order to eliminate a single point of failure. Now keep in mind that if just mount both instances data stores to the same SAN then you still have a single point of failure and you may as well go with a shared data store.
No matter what clustering option you select you will still be able to use replication and reverse replication. The replications services implement a cluster aware interface that ensures that they only run on the master node of the cluster.