I am surprised to read about the changes in the Exchange2013 Architecture. Microsoft has changed the architecture again and now using the combo approach which means running multiple roles in a single role or a single install. It was not too far when Microsoft’s experts were suggesting using the multiple role deployment even in Exchange2010 to make the use of available hardware but didn’t know that will bring the change in the new release of Exchange2013. So this what it has in the new architecture
Out of new changes there is BIG SURPRISE that there are only TWO SERVER ROLES, MAILBOX ROLES AND CAS ROLE. Microsoft has combined other server roles into these two roles. So you can install a HUB TRANSPORT and UNIFIED MESSAGING separately anymore.
There are only two roles in Exchange 2013 i.e. Mailbox and Client Access Server which means no Hub Transport server and there is no news for Edge Transport server but Exchange 2010 Edge can be used.
Exchange2013 Mailbox Server contains all the roles yes it contains all the Client Access Protocols, Hub Transport Services, Mailbox Database and Unified Messaging.
The CAS Server itself does not do anything data rendering, it is just a thin layer which does proxy, redirection and authentication. It is a stateless server. It does offer all the protocols, HTTP, POP and IMAP and SMTP.
RPC is no longer a supported direct access protocol. Means all Outlook connectivity must use RPC/HTTPS (Outlook anywhere). Does it mean goodbye to MAPI?
Outlook clients no longer connect to a server FQDN as they have done in all previous versions of Exchange. Outlook uses AutoDiscover to create a new connection point comprised of mailbox GUID, @ symbol, and UPN suffix.
The unit of high availability is still the database availability group (DAG). The DAG still uses Windows 2008 clustering. Continuous replication still supports both file mode and block mode replication. However, there have been some improvements. Failover times have been reduced as a result of transaction log code improvements and deeper checkpoint on the passive nodes.
The Exchange Store service has been re-written in managed Now, each database runs under its own process, allowing for isolation of store issues to a single database.
New Architecture benefits
Version upgrades flexibility No more rigid upgrade requirements. A Client Access server can be upgraded independently and in any order in relation to the Mailbox server.
Geo-flexibility Because all the processing and data transformation takes place on the Mailbox server, we’re no longer constrained to having both a Client Access server and a Mailbox server in each site. You can now choose to have a central Client Access site for all protocol traffic if you want.
Session indifference With Exchange 2010, session affinity to the Client Access server role was required for several protocols. In Exchange 2013 Preview, the client access and mailbox components reside on the same Mailbox server. Because the Client Access server isn’t doing any data rendering, we only require layer 4 load balancing. Layer 4 load balancing is protocol- unaware and balances traffic based on IP address and TCP/UDP port. Does it mean now you can use WNLB without the issue of IP Affinity? And you don’t need high-end load balancers anymore?
Deployment simplicity With an Exchange 2010 site-resilient design, you needed up to eight different namespaces. With Exchange 2013 Preview, the minimum number of namespaces drops to two. If you’re coexisting with Exchange 2007, you still need to create a legacy hostname, but if you’re coexisting with Exchange 2010 or you’re installing a new Exchange 2013 Preview organization, the minimum number of namespaces you need is two: one for client protocols and one for Autodiscover. You may also need an SMTP namespace. So there are three names space in total.