Today’s internet traffic typically comprises data, voice, or video traffic with no extraordinary means to logically segregate traffic based on its network latency sensitivity. Applications have varying requirements for bandwidth, latency or jitter. Some apps require high bandwidth, such as large file downloads or video traffic, and certain apps require low latency, such as online gaming traffic or high frequency trading. The online gaming industry is on a rapid growth path and has become an exciting mainstream revenue source. With an increasing demographic that streams gameplays, the cloud gaming services are bringing new online gaming experience closer to the consumers and it will require support from the 10G initiative driven MSOs to deliver the best quality of experience by ensuring high bandwidth and low latency or low lag support. By enhancing the user experience, the MSOs have an opportunity to generate a new revenue stream. Welcome to the world of Low Latency DOCSIS! The LLD architecture as proposed by CableLabs enables a logical separation of the latency sensitive nonqueue building traffic and regular queue-building internet traffic in to two separate queues. The two queues, Low Latency SF and Classic SF are encapsulated in an Aggregate Service Flow (ASF) to shape the traffic. A key innovation that is part of the LLD architecture is a new scheduling service known as Proactive Grant Scheduling (PGS).
There are multiple sources of latency in DOCSIS networks, including protocol/application dependent queuing delays, propagation delay, Request-Grant delay, channel configuration (OFDM or SC-QAM interleavers, cyclic prefix, FEC, etc.), and switching/forwarding delays. The purpose of LLD is to reduce latency from two of these sources – protocol/application dependent queuing delays and Request-Grant delays.
This paper will focus on the LLD architecture basics and experimental results from the lab studies using the concept of an LLD ASF and PGS in the DOCSIS Upstream. The paper will also compare LLD capable system latency with classic latency.