ANNOUNCEMENT - The community will be down for maintenace this Thursday August 13 from 12:00 AM PT to 02:00 AM PT. As a precaution save your work.
Showing results for 
Search instead for 
Did you mean: 
Cisco Employee

How distributed is a distributed EPC?

The recent Unstrung article ( has triggered quite some discussion regarding the level of distribution we can expect in future mobile architectures. But the question is, "how distributed is distributed?" Is the vision to have the IP Point of Attachment (Mobile GW) co-located at the base station, much the way folks are talking about local breakout with femto-cells? Or is the vision a little more conservative?

The first thing to recognize is that today's packet core architectures are "super-concentrated", with the IP Point of Attachment (e.g., GGSN) being located in a centralized data center. Certainly its not uncommon for a very large operator (lets say 80M subs) to have a very few data centers (lets say 4); or 20M subs per location. Contrast this with legacy MSC sites which typically serve between 250k and 1M subs and we see a clear disparity, and hence the "super-concentrated" description.

The next area to consider is sustained throughput. In an upcoming Heavy Reading video webinar ( I discuss the cell site depreciation for a 10 GB/month user - the hypothesis that HSPA is fine for today's 1GB/month iphone user but that LTE and EPC needs to have ambition to support more than iphone, for example to scale to support the 10 GB/month user. A single 10GB/month user in this example consumes something like 75 kbps in the busy hour. So looking at our "super concentrated" deployment, would require a massive 1.5Tbps of sustained throughput across the mobile GW! Contrast this with an IP Point of Attachment at the MSC site, and it only have to support 75 Gbps of sustained throughput - a figure that iometrix concluded ( could be supported by Cisco's GGSN today!

Then we need to look at the flows. Unfortunately, the reality is that the majority of flows may be going off-net and so need to access the Internet. Outside of the US, operators (and often countries) have centralized IP POP architectures where peering and transit services can be accessed. So we need to be cautious about any predicted cost savings from distributing the IP Point of Attachment but still requiring the flows to be transported to a centralized IP POP for connection to the Internet.

Then we need to recognize that LTE will likely be deployed in islands with frequent inter-RAT handover between LTE, 3G and perhaps even 2G technology. These legacy systems have hierarchical access networks, typically terminating at the MSC-site. So if frequent handovers need to be supported, any benefits from distributing the IP Point of Attachment will be lost if the user is accessing via the legacy RAN network.

Finally, from an organizational perspective, distributing the IP Point of Attachment will likely involve hosting such functionality on a transport platform. We need to recognize that many operators have different organizations dealing with packet core and transport and so even when there is a clear architectural benefit, there may be some organizational inhibitors to fully distributing the IP Point of Attachment.

Bottom line, operators need to start planning today to evolve their "super concentrated" packet cores towards a "more distributed" architecture. That architecture will surely see mobile GWs move out of the centralized datacenter and head towards the MSC-location. More distributed approaches may apply to niche scenarios, for example to greenfield LTE deployments or where a country's regulator has already mandated a very distributed IP POP architecture.

Everyone's tags (2)
Cisco Employee

Re: How distributed is a distributed EPC?

In the cost aspect, following factors play a role:

  • Cost of Backhauling yourself vs. cost of ISP to backhaul - and this becomes especially important for countries where they have "concentrated" Internet PoPs. Also depends on the mobile operator, if it has enough capability (wired infra, skills et all) to backhaul 10s of 100s of Gigs.
  • Distribution has extra CapEx cost and this gets ignored. Especially the capex involved in PCEF (Policy Control Enforcement Point), Monitoring equipment, security, Internet routers et all
  • Though, we have to be careful that CapEx is not completely removed when we move from distribution to centralization, its only "partially" reduced.
  • Opex cost of distribution - engineering and operations cost increase with the number of locations
  • There is a cost associated with centralization, this again gets ignored. when you start to backhaul 100s of Gbps or even Tbps, all the network elements like core routers, datacenters switches need to get upgraded with high speed linecards.
Cisco Employee

Re: How distributed is a distributed EPC?

One pitfall that I have seen with distribution is that operators focus on the very busy markets, and try to break that market away. This is right strategy, but only partially.

Distribution should account for a a national view of all markets. And this will create a more regional distribution structure - neither too centralized (like it exists in most operators today) nor too distributed.

Content for Community-Ad

This widget could not be displayed.