MULTI ACCESS EDGE COMPUTING (MEC) OVERVIEW

I am learning about 5G(and telecom in general) and writing about it helps me walk through it better. Instead of directly introducing MEC as a concept, let's introduce it as a solution to a problem and its challenges.

The problem:

We need high speed low latency access to data. The packets in the mobile network are transferred from the user device to a local eNodeB(the tower), then its forwarded to the regional S gateway, which is generally 100s of kms away in a datacenter and then these packets are forwarded to appropriate P gateways which are then released to the public internet or IMS, voice call handler. A few such datacenters can handle a whole country wide network.

The main latency in a mobile network comes from forwarding these packets from eNodeB to the datacenter where they are then segregated into voice or internet or various other packets and forwarded. Each stage of forwarding, be it from UE, eNodeB, SGW or further requires removing previous stage's header and adding its own header preserving the internal data which also takes more time, unlike home internet for example which directly pushes packets into the public internet, this can be observed with a simple ping on a home wifi vs mobile network as shown.

Wifi:

4G Jio:

Low latency applications like gaming or self driving tech will suffer significant setback with such (5x slower, even in a city) network.

The Proposed Solution

The straightforward solution we see is to bring the gateways to the internet closer to user. MEC does exactly this, every few regional eNodeBs can be connected to a local server or containerized(literal shipping container) server which can serve the regional users very fast and is directly connected to the internet with fiber. We also get the added advantage of distributed high tolerance system and failures are localized. We can also maintain a local cache of the data like high quality Youtube videos and give high QoS to users. Self driving or AR tech companies can also install local hardware in these servers to enhance their services with local AI support.

The <What are you waiting for part>

MEC solves the big problem of latency but introduces a new problem which did not exist before. In a traditional network, when a user moves from one eNodeB to the other, the app being used by the user, the buffer, the settings all are at the core of the network and a simple handover will let the user's device continue its data stream, but a localized solution means its app is running on the regional server and handing over will require the migration of full current instance of the user data to the next node. This needs application of strategies like Live migration of VMs, typically pre migration techniques which maintains the data image of the user on multiple servers around the current one and just keeps copying the dirty pages in the memory and once the user switches, the vm goes down for a few milliseconds and transfers the remaining data to the new server. This will also be particularly challenging as the future 5G tech has very low range, meaning frequent handovers . This part is not clearly implemented or defined yet.

The second problem is that different MECs means different capabilities for different regions, this means the eNodeBs will have to advertise their services, like whether it supports self driving kind of applications etc, so that the user can make a decision.

Any discussion is welcome.