Microsoft Exchange Unified Messaging Architecture Considerations

The question that comes up more often than not with customers is how if possible can I make a remote site survivable for voice mail? There are some things that need to be considered before a decision like this is made. What does survivable mean in terms of the budget, resources and administration? Before we go down that road we need to talk about the functionality differences the user experience.

There are 3 major architecture considerations for unified messaging with OCS or a Gateway. Centralized,
Distributed UM Server Deployment, Distributed UM and Mailbox Deployment and each have unique affects on the user's experience. The problem with the last two options is the introduction of MAPI across a WAN. The exchange servers talk to each other using MAIP for certain requests. The Unified Messaging Server uses MAPI to get mailbox information and greeting information out of the user's mailbox.

When the Unified Messaging server uses MAIP protocol to retrieve user content the caller experience becomes degraded. If the round trip time from the Unified Messaging server to the Mailbox server is greater than or equal to 100 milliseconds the user's personal greeting will never be played. This is because the Unified Messaging server times out when trying to generate a temporary copy of the greeting stored file in the user's mailbox. Once the Unified Messaging server times out it attempts to play a copy of the users recorded name stored in the local sites DC. If that times out the Unified Messaging server will attempt to speak the user's name.

Centralized Architecture

The centralized architecture eliminates the possibility of having a bad users experience because of the latency on the WAN with MAPI. This solution does not support remote site survivability. If the WAN goes down the user will either hear fast bust tone from the OCS applications or dead air from the gateway. This solution relies on the WAN being up and the usage of SIP, RTP/SIP between the remote site and the Unified Messaging server in the data center. The centralized model eliminates the need for Unified Messaging servers and Mailbox servers in the remote sites. This also creates a centralized location for administration for both servers and application.

With the centralized deployment all users will get their personal greeting and Out of Office messages played to the callers. The further in milliseconds a user is away from the unified messaging server, the longer it takes for the ringing to stop and the auto attendant or subscriber access to answer. Once the subscriber access or auto attendant answers everything is local and the menu response times are very fast.

Distributed UM Server Deployment

The distributed Unified Messaging server Deployment at first glance looks good to most customers. The problem with this type of configurations is the introductions of MAPI across the WAN. In this configuration the site becomes survivable in the fact that all calls will be answered in some form by the server. However if the distance from the Unified Messaging server is greater than or equal to 100 Milliseconds the user's personal greeting will never be played. Also in this Configuration subscriber access and auto attendant directory transfer becomes increasingly slow. As the network latency increases so does the delay to log into a local subscriber access number. This is evident by the auditable hour glass sound Unified Messaging plays during the authentication process. To put this into perspective a users trying to long into outlook voice access across a T1 at 100 Millisecond delay, no jitter or packet loss, will have a 9.5 – 10 second wait until they hear their first voice mail.

This option is a common misconception for customers because architecturally it makes sense, however functionally it is worst out of the three. No network is perfect especially across the WAN and if personal greetings for users are a business requirement then this option is not for you.

Distributed UM and Mailbox Deployment

This option requires that all the Microsoft Exchange server roles are deployed in each location. This option is polar opposite from the centralized model which many customers are moving towards. The distributed architecture becomes a consideration if the business requirement of remote site survivability and low delay are mandatory. With this configuration there is still the possibility of a MAIP across the WAN scenario. If a caller contacts a local auto attendant and performs a directory-look-up to a user that is in another site, the system will attempt to play the personal greeting using the MAPI protocol across the WAN. This will time out and the personal greeting will never be played. Out of the three options this is the most expensive and administratively intense. There can be large mailbox storage, CAS, HUB and Mailbox redundancy requirements. Additionally to the servers an internet presence for the remote locations may be required to allow users to access their mailboxes through Outlook Web Access. This last option tends to be the best fit for Global regional data centers. This scenario would still employ the remote sites to SIP across the WAN for voice mail however users look up could allow MAPI across the WAN for voice mail transfer