Article

TCP vs UDP – The Fight for Space On A Shared Network

In our daily life we have come to rely on mobile internet access – as a reliable, fast, shared resource. What you may not realise is that underneath the layers of content & streaming access there are two main transport protocols that communicate between device and the internet. Stripping away the layers of radio transmission and content encapsulation from applications the fundamental transport of data packets is competing between TCP – transport control protocol and UDP – user datagram protocol. These two types represent, at a practical level, different forms of communication, connection-oriented and connectionless respectively.

 

TCP relies on acknowledgment between sender and receiver of the content transmitted as well as the quantity of that content at any one time – at each stage of the many waypoints between your mobile device and the web servers. TCP is labelled as a connection-oriented protocol.

UDP is more of a per packet-based acknowledgment between sender and receiver. Many packets can be sent without acknowledgment and may go by different routes between mobile device and content server. UDP is labelled as a connectionless protocol.

The former, TCP, has been the backbone of internet access for years and the latter, UDP, has been more adopted recently by large streaming vendors in the form of QUIC protocol. It is the reliability of the mobile network that has made UDP more practical with applications assuming that their packets get through most of the time. TCP on the other hand builds in additional reliability by acknowledging packets and re-transmitting unreceived packets automatically.

In mobile networks TCP is about 55% of traffic and UDP 45%* – which illustrates the rapid adoption of UDP over the last 5 years.

These two protocols live in the same network but have different characteristics of behavior. With TCP the network owner can have an effect on both the amount data exchanged and bandwidth; In UDP the relationship is between handset and content server only with the network acting only as a conduit. These protocols are mainstream and heavily adopted but in effect compete for the same shared resource, the mobile network.

What are the possible effects of competing protocols? As UDP is between application on device and server it can initially send/receive as many packets as possible without considering the available bandwidth on the network. TCP on the other hand recalibrates the amount of data it sends, reliably, on the available bandwidth – building to a maximum level. In this case TCP could be considered a ‘polite’ protocol than handshakes on how much can be sent and received regularly during the connection. UDP could be considered as the impolite protocol that uses as much as needed and only is concerned if packets are lost and not received. In this respect TCP could be backing off in face of bandwidth used for UDP.

Fortunately for TCP there are mechanisms to help it, and network owners, stopping UDP from hogging too much of available resources. Enea Traffic Management team specialize in these capabilities and the benefits are surprising:

#1 TCP startup & ramp up: TCP negotiates the quantity of data sent over the network link at the start of the session, and then increases it as packets are reliably transferred. Using more aggressive settings, the startup phase can reach the maximum bandwidth much more quickly.

#2 TCP ramp down:This is where TCP reduces the quantity of data if it gets too many unacknowledged packets. This can happen if there is a reduction in bandwidth due to a change in the network conditions or if other traffic flows, such as UDP, is hogging too much of the network. The amount of backoff can be better balanced to be more aggressive to UDP for the same device (to ensure the TCP based application is not starved of connectivity) or, in the case of radio change to decrease more rapidly. 

#3 Consistent Connectivity Probing:  By checking whether the user has changed radio (4G to 5G or vice-versa) or is facing congestion in the mobile network. TCP can adapt its reliable transport more quickly than the apps using UDP as it is network aware and constantly measuring round trip time directly.

These three key areas have material effect on the time taken to download / access larger content. The average web page size for mobile sites is 2 MB and getting that content reliably and quickly to the user matters.

TCP optimization can generate 10-25% faster downloads if tuned correctly; our experience shows this for 4G profiles (e.g. 10-40 Mbps) to 5G (e.g. 80-600Mbps). The effect of this can be seen in user tests and drive tests particularly as networks compete for top spot regionally.

An underlying benefit is the additional network control it gives for (OSI) layer 4 (transport) control, ensuring that TCP gets its fair share of bandwidth in the face of UDP. In effect having TCP be more aggressive but still reliable gives more balance for the transports that mobile devices depend on for their data in a shared network.

It is also worth noting that TCP acceleration in variable radio environments (4G, 5G, 5G-NR) is key as coverage is not universal. Automatically detecting changes and adapting means that the user’s available bandwidth is always used to the maximum and in control of the telecom network.

 

Check with the Enea Team today on the value and insight that TCP acceleration can bring and our techniques for managing UDP and video streaming with Enea Traffic Management!

*Data from Enea deployments in 25 countries, 2024.

Related insights

Omdia: Manage & Monetize Data Access: Best in Class Intelligent User Plane Solutions

Read more

Tags: Encrypted Video, Network DPI, TCPAcceleration, Traffic Management

Virtual Private Network (VPN) & Mobile Network Access

Read more

Tags: Carrier Wi-Fi, Network DPI, signaling security, Traffic Management

CyberSecurity Month: Parents & Network Safeguards For Their Kids

Read more

Tags: 5G, Deep Packet Inspection, DPI Filter

Meta & Telco CSP – A New Status Quo?

Read more

Tags: Traffic Management