Tuesday, October 18, 2011

Modeling Policy Server Scale and ROI

     
A month ago I covered the deployment of policy management by Bharti-Airtel (the largest MNO in India with over 170M subscribers). Bharti's IT director said (here) then that "..The challenge that we have is that our scale is huge".

While most networks are significantly smaller, policy management scaling is an issue vendors and operators need to handle. In addition to the growth in data consumption we also see new use-cases and M2M applications (here, here) that will add load to policy servers.

Scaling was addressed by policy server vendors as a competitive advantage for some time now (BroadHop, Bridgewater). Diameter routing/load balancing technology and products (here) were added in recent months to help policy management to scale.

However, something was missing - what are the expected performance requirements?

A new white paper by Graham Finnie (pictured), Chief Analyst, Heavy Reading. (commissioned by BroadHop) builds a model to help set performance goals: "In the past two years, policy management caught fire as network operators sought better ways to manage the way bandwidth is allocated and congestion is handled. Now, many are looking to move on from these early deployments, and seeking to put policy at the heart of their traffic management and service development strategies. But as policy deployments scale up, it raises major new issues for operators. Can policy servers cope as new use cases are added? What will it cost? And can the new business case really stack up?"

Conclusions:
  • A 3GPP network of 10 million mobile subscribers will scale, based on our assumptions, from handling about 2,200 TPS in the Base Case to almost 23,000 TPS in the Mature Case, which is assumed to be after three years.
     
  • We modeled the TPS impact of beginning the transition to an all-LTE network. This also assumes a network with 10 million subscribers, in which the customers are gradually transitioning to LTE, and that during this transition, wireline customers are also transferred to the same policy environment. In this case, policy scales from nearly 25,000 TPS at the end of Year 1 to more than 75,000 TPS at the end of Year 3.
     
  • the cost of policy is only a small proportion (around 2 percent) of the overall cost of a new LTE build over a five-year build period in the next-generation policy case – and given the high strategic value of policy, this suggests to us that the return on investment (ROI) is likely to be positive in the short to medium term
See press release "BroadHop and Heavy Reading Study Finds Policy Servers Must Scale Up Massively as Mobile Operators Move into the LTE Era" - here ; White paper - here.

1 comment:

  1. So the TPS is very high, then what can PCRF vendors do? What can service providers do?

    It doesn't say in the paper but seems to suggest to use DRA, or distribute the policy decisions to PCEF, which are not of Broadhop's best interest.

    ReplyDelete