Four application groups have clear low-latency requirements: first, financial and electronic trading users, especially electronic traders engaged in high-frequency trading (HFT); second, TCP-based high-definition video services, including 4K/8K live streaming and on-demand, high-definition video conferencing, and future virtual reality (VR) services that demand both high bandwidth and low latency; third, certain cloud services, notably virtual machine migration, hot data backup, and latency-sensitive cloud desktop and cloud payment services; fourth, 5G mobile network transport bearer services, where current 5G networks reserve very strict latency targets for the transport bearer layer and require new low-latency transmission techniques to meet them.
Extreme latency requirements of financial and trading applications
Information asymmetry drives trading profits: lower latency means you receive information sooner and your orders reach exchanges earlier, increasing the chance of profit. In developed financial and trading markets (especially the United States), high-frequency trading (HFT), also called algorithmic or machine trading, has been established for years, covering futures, equities, foreign exchange, and other markets. Statistics indicate that HFT accounted for 61% of total trading volume in the United States in 2009, rising to 70% in 2012, and in the UK it reached 77% in 2011.
As early as 2008, the consulting firm TABB Group published a study titled "The Value of a Millisecond: Finding the Optimal Speed of a Trading Infrastructure," noting that for a trading firm in the US electronic markets, being 5 milliseconds slower in system processing (including transmission latency) than competitors can reduce profit by 1%, while being 10 milliseconds slower can reduce profit by 10%.
Figure 1: Delay performance of a microwave relay circuit between New York and Chicago
The adage "time is money" is particularly literal in HFT. An InformationWeek report titled "Wall Street's Quest To Process Data At The Speed Of Light" claimed that in the US electronic financial markets, a 1 ms latency advantage can be worth $100 million. A 2014 pre-IPO disclosure from a US HFT firm showed that over 1,278 trading days it reported losses on only one day.
HFT has driven trading firms to pursue minimal latency: servers are colocated as close as possible to exchange servers (NYSE, NASDAQ, CME, etc.), ideally within the same data center, and extreme efforts are made to minimize transmission circuit latency.
High throughput needs for 4K/8K and VR services
TCP is the dominant transport on the Internet. Its acknowledgment mechanisms ensure reliability but can limit throughput. TCP throughput is constrained by three factors: bandwidth (BW), round-trip time (RTT), and packet loss rate (ρ). Assuming sufficient bandwidth and negligible packet loss in a well-managed network, latency becomes the decisive factor. When latency is large, users cannot realize higher throughput by increasing bandwidth alone, a situation sometimes called a "bandwidth black hole."
TCP's congestion window (CWND) field is 16 bits, so the maximum CWND size is 64 KB (65,536 Bytes). With a typical MSS (maximum segment size) of 1460 Bytes and negligible packet loss, latency remains the critical factor. For example, at 10 Gbps bandwidth and 10 ms one-way delay (20 ms RTT), TCP's maximum throughput is only about 26.3 Mbps, far below the network bandwidth.
Industry practice often assumes that real-time, high-throughput services such as 4K/8K video require an effective throughput about 1.5 times the actual bitstream to ensure quality. Therefore, a 4K video may need 30–45 Mbps. Using the above constraints, the allowable RTT for such throughput is roughly 12–17 ms.
There are proposals to address TCP window limitations, for example extending the sliding window size (RFC extensions) or using UDP at the application layer or multiple TCP threads. However, these require broad network upgrades and are hard to deploy quickly. Within existing networks, reducing latency is the most direct and effective way to mitigate TCP throughput limits. IDC and CDN deployment should account for the latency needs of 4K/8K and other high-throughput services and be planned accordingly.
Latency needs of real-time cloud services
The rise of cloud computing, big data, and the Internet of Things has moved many services to the cloud. Data centers, as the physical carriers of the cloud, are becoming central to network traffic, and migration of data within and between data centers is increasingly frequent. This trend, which we call cloud communications, includes some services with strict real-time requirements and therefore low-latency demands.
The most typical real-time cloud task is virtual machine migration. For example, live migration often requires latency below 10 ms. Cloud data hot-backup, disaster recovery, and large-scale collaborative computing also impose strict latency requirements.
As more upper-layer services migrate to the cloud, cloud transport networks must meet tight latency targets to preserve user experience. For instance, optimal cloud payment experience requires latency below 10 ms, while cloud desktop services require latency below 20 ms for the best experience.
5G mobile network transport latency requirements
5G is at an early stage and proposes ambitious goals. ITU named the 5G standard IMT-2020. China's IMT-2020 (5G) task group released a white paper in 2014 titled "5G Vision and Requirements," which lists key 5G capabilities including user-experienced rates of 0.1–1 Gbps, connection density of one million devices per square kilometer, end-to-end latency on the order of milliseconds, traffic density of tens of Tbps per square kilometer, mobility over 500 km/h, and peak rates of tens of Gbps, as shown in Figure 2. The white paper identifies user-experienced rate, connection density, and latency as three fundamental 5G performance indicators.
Figure 2: IMT-2020 (5G) key capability indicators defined by ITU-R M.2083
Compared with 4G/LTE, 5G sets more stringent latency requirements. Going forward, wireless and transport network experts must collaborate closely to define clear and practical latency targets to support 5G evolution.
In summary, with the development of electronic trading, high-definition video, cloud computing, and 5G, latency has become a critical performance metric for communications networks. Low latency will be an important competitive capability for network operators. In developed and international leased-line markets, low-latency circuits are already a distinct product category, and latency is a key SLA and differentiation metric for service providers.
For example, the undersea cable operator PACNET, owned by Telstra, classifies its cable transport service between nodes into Low Latency, Standard Latency, and Best Effort categories with differential pricing. TATA Communications announced in 2012 the construction of a global low-latency network targeted at financial firms and HFT. Verizon established a financial services division and launched a dedicated low-latency network connecting major financial and trading data centers. The dedicated circuit between CME's Cermak data center in Chicago and the Carteret data center in New Jersey had a round-trip latency of 14.5 ms, about 40% lower than standard circuits.
