Unified Communications Server Architecture: Location Matters
Cisco has consolidated our unified communications server architecture twice since we first transitioned from PBX systems. Initially, we deployed Cisco Unified Communications Managers in all 256 global Cisco offices.
The first big shift came when we moved to a centralized Unified Communications architecture, deploying 15 Cisco Unified Communications Manager clusters worldwide, which support more than 130,000 hardware phones and 70,000 software phones. The centralized clusters deliver voice services to other Cisco offices over the Cisco global WAN.
The enabler for our centralized architecture is a feature in Cisco branch office routers called Survivable Remote Site Telephony (SRST). If the WAN link between the router and the Unified Communications Manager cluster goes down, calls to and from that office automatically travel over the public switched telephone network, so there’s no loss of service.
In general, we deploy Cisco Unified Communications Manager clusters where our employees are:
- Our largest campuses, including San Jose, California, Bangalore, India, and RTP, North Carolina each have their own clusters.
- Other clusters handle all offices in a geographic region. We have one cluster for the western United States, one for the eastern U.S. plus South America, and another for most of Europe, the Middle East, and Africa and several smaller clusters in the Asia Pacific theater.
- Cisco IT built two dedicated clusters for the voice portion of Cisco WebEx conferences, and another dedicated cluster for Cisco TelePresence sessions. There’s nothing different about Cisco TelePresence that necessarily requires a separate cluster, but we decided to keep it separate for manageability. The Cisco TelePresence cluster is split between two data centers about 20 miles apart, for availability reasons.
- Finally, we have a few smaller clusters for geographically isolated Cisco offices where latency might otherwise be an issue, such as South Africa. We have others in countries with special regulations for signaling or records retention.
Every cluster includes its own Cisco Unified Communications components, including subscriber nodes, publisher, and so on. In 2008, we added a pair of Cisco Unified Presence Servers to every cluster. Presence is a very popular feature here at Cisco because it lets us see whether our coworkers are currently on the phone before taking the time to dial. Cisco employees save from 10 to 15 minutes a day up to an hour or more because of presence.
The Cisco Unified Presence Servers are not counted in the maximum number of servers allowed per cluster, so could theoretically add it everywhere. But a couple of data centers didn’t have enough physical space for a pair of Unified Presence Servers, so we implemented them in another location on the same metropolitan area network.
The next big evolution in our server strategy is coming later in 2010, when we’ll virtualize most of our Unified Communications Manager and Unified Presence Servers and host them on the Cisco Unified Computing System (UCS). We expect the unified computing platform to improve our already excellent performance and reliability. And, we’ll only need about one-eighth the number of servers, reducing management requirements and making the data center greener.