This article originally appeared in the May 2012 issue of INTERNET TELEPHONY magazine.
While carrier service providers struggle with the transition from 3G to 4G, enterprise organizations are slogging through similar challenges as communications converge toward a unified voice and data model. The de facto standardization of unified communication systems on SIP to provide voice services is a foregone conclusion. Managing the resulting load on data center infrastructure is all that’s really left for organizations to get a handle on.
Organizations are transitioning simultaneously from static data center to dynamic data center models. Surveys show a consistent trend toward virtualization of the data center, with anywhere between 38 and 45 percent of organizations having adopted server virtualization already. The trend toward utility, cloud-computing style infrastructure is undeniable. The lure of a vast pool of compute resources, available for allocation on-demand, is strong. Idle, underutilized compute – once the mainstay of the data center to assure availability and performance – is a thing of the past. Efficiency is king, and its chancellor is virtualization.
Capacity planning, however, is not a thing of the past. In many organizations capacity planning strategies have been outflanked by the convergence of UCS and virtualization simultaneously. The dynamic consumption of compute required to support a high-quality, SIP-based communication system moves the burden of managing capacity from a pre-deployment exercise to a real-time fire drill.
SIP server capacity is impacted by a variety of factors, all of which are highly variable. The role-focused nature of SIP supporting infrastructure complicates the process, as different roles also have different compute and network resource consumption profiles, and these are also dependent on real-time factors such as current load, network conditions, and length of calls occurring.
Integration across the data center with authentication systems, address lookups (DNS), processing, and state management must all be taken into consideration, especially when virtualization is involved. Each role has a unique resource profile, with some components such as B2BUA servers imposing a higher processing load on the virtual server. Stateful components also have higher real memory requirements (30K per concurrent call is currently suggested as a starting point) to avoid the impact of virtual memory on performance as call state and other SIP-related data is moved back and forth between disk and real memory.
Complicating the provisioning of UCS components is virtual networking. Provisioning resources for a given SIP function may be as easy as firing up a golden image, but where it’s placed within the network may have a profound impact on its performance.
Virtual networking changes traffic patterns, shifting from a primarily north-south pattern to an east-west pattern. Traffic flows between virtual servers, traversing a complex network of virtual adapters, switches, networks and infrastructure as well as through traditional network infrastructure. Provisioning latency-sensitive services such as VoIP in such an environment can be challenging.
The notion that a-few-sizes-fit-all inherent in utility computing models such as cloud makes this challenge more difficult. Such models are based on the assumption of more static capacity profiles than is true for SIP-based communications. Capacity of web services is more easily managed based on easy to capture metrics such as concurrent connections, and thresholds based on such measurements can be easily implemented.
Real-time unified communications, however, with multiple roles and varying demands per user based on activity and length of call and unknowable conditions such as network congestion and latency, must be managed based on real-time metrics with resources dynamically provisioned. The capacity of a virtual server within the SIP architecture may suffice to provide acceptable call quality for 10 users today, but it may not manage that success tomorrow for even four.
A more dynamic, intelligent system of not just provisioning but monitoring quality and conditions must be in place to ensure quality of unified communications. While virtualization affords the opportunity to mitigate quality issues arising from resource constraints through on-demand provisioning, such systems cannot do so without real-time, actionable data upon which to base such decisions. A holistic view of the entire system – of all components (roles) comprising unified communications infrastructure – is required to ensure no single service within the system becomes a bottleneck and impedes performance, and thus call quality.
Virtualization is, for UCS, a double-edged sword. It offers both a solution to quality issues while introducing additional challenges that must be addressed to ensure the success of a virtualized UCS implementation.
Edited by Stefania Viscusi