Note: This information was obtained from a conversation and is therefore to be understood as an unfiltered transcript.
Das Exchange 2013 Environment besteht aus der „All-in-One“ Mailbox Server Rolle und der CAS Rolle, der aber wenig mit der 2010 CAS Rolle gemein hat. Die CAS Rolle im Sinne von 2010 liegt auch auf dem Mailbox Server, welche nun alle Rollen innehält.
The reason for this is the current factors in the IT industry: CPU performance is cheap and not a decisive factor and people want a simple, standardized and, above all, scalable solution. On top of this, fault isolation is becoming easier and the use of today's hardware more effective. Especially with virtualization. [1]
An EDGE Transport Server does not exist at the moment, if needed/considered a 2010 Edge Transport can be integrated, such as described here.
Die CAS Rolle übernimmt keine Client verwaltenden Rollenfunktionen außer Authentifizierung, Proxying Dienst und eine begrenzte Redirection Funktionalität. Die CAS Funktionalität eignet sich wunderbar in Failover Situationen um den Outlook Clients im Cached Mode nach wie vor eine Verbindung zu Exchange zu suggerieren, während der Zeit die das Failover auf die andere Site läuft. Die „Session Affinity“ sitzt somit nicht mehr beim Load-Balancer sondern wird vom CAS geliefert. Details können read here where the behavior of the new CAS is explained in detail. Layer-7 load balancers are therefore no longer required; layer-4 load balancers are recommended.
Interesting information on the required environment variables can be found at here.
A comparison of an earlier Exchange 2010 organization with the current 2013 environment:
2010 Architecture:
2010 wurde ab ca. 2005 entwickelt, dabei war Multicore/Processing noch nicht so verfügbar wie heute, somit wurden die Rollen sehr geteilt. Es gab damals viele Gründe für „Loosely connected“ Server. Größte Herausforderung stellt bei 2010 das Load-Balancing dar. Layer7 LB wurde benötigt. Viele Namespaces werden benötigt.
With its all-in-one role, 2013 offers the possibility to easily scale from small to large using the modular principle. Update scenarios have been greatly simplified as different update versions work together. Each server is self-contained and only communicates with each other via EWS, MRS, SMTP etc.; cross-communication between the components has been prevented.
Thanks to the new CAS role, only Layer4 LB is required.
A DAG can consist of 16 mailbox servers, the DAG members can be located in different data centers, which are networked with a broadband connection.
With regard to hardware sizing, there is currently no calculator or similar. However, knowing that all roles run on one and the same machine, the current Calculator for Exchange 2010:
The division of Store.exe into service and worker processes offers performance advantages; each DB has its own worker process. This saves performance when replicating passive databases and increases the speed when switching from passive to active. The IOPS have been further reduced.
For multiple mailbox servers, we recommend an even load of databases per server. Ideally, a database should carry 1.67 TB of user data and be 2 TB in size. Several passive databases distributed over several servers allow a server to be restored more quickly, as seeding now takes place in parallel from several servers. Each server transfers database data at approx. 20MB/s to another server during recovery. Scaling is therefore crucial in the event of server failures or database failures and restores.
In this case, access to DB1 on MBX15-1 fails, either because the database does not respond or the OWA service hangs, which is automatically detected by the server via probes and responders. The error is recognized and the service is restarted automatically. The service runs again, but shortly afterwards the service may stop again. The service could no longer be restarted automatically. The Managed Availability Service notices this and switches off DB1 on MBX15-1 and switches DB1 to MBX15-2 is active. The clients are redirected to the new server.
A screencast with the most important architectural changes can be found on Technet:
The news is also discussed in another video. I think both videos are worth watching.
http://technet.microsoft.com/en-us/video/whats-new-in-exchange-2013.aspx