Quality of Service (QoS)

The time it takes for a packet to travel from the edge cache to the user equipment. In proactive caching, this is the primary metric, as the goal is to serve content with sub-millisecond delay.

• Target: Often < 1 ms for URLLC-like cached data in 5G.
• Measurement: Calculated using timestamps in real-time transport protocol headers.
• Impact: High latency negates the benefit of edge caching, making the pre-fetching strategy ineffective.

The variation in latency between packets in a single flow. For cached video segments or tile-based 360-degree video, high jitter causes bufferbloat and stuttering.

• Mitigation: A jitter buffer at the client smooths playback by queueing packets.
• QoS Mechanism: Traffic shaping with a token bucket algorithm polices the burst rate to ensure a smooth, consistent packet stream from the cache.

The percentage of packets that fail to reach the destination. In a MEC caching scenario, packet loss on the radio link requires retransmission, which destroys the low-latency advantage.

• Cause: Congestion on the backhaul or poor radio conditions.
• QoS Solution: Backhaul offloading via local caching bypasses congested core network links, directly reducing loss. Forward error correction is applied to cached data streams.

The minimum data rate guaranteed for a specific cached content flow. This is a critical Service Level Agreement (SLA) metric for network slicing.

• Application: A 4K video cache requires a guaranteed bit rate of ~25 Mbps.
• Enforcement: The token bucket algorithm (CIR/PBS) meters traffic, ensuring a slice dedicated to cached video delivery always gets its committed rate, even under network congestion.

A subjective quality metric rated by users on a scale of 1 (bad) to 5 (excellent). For proactive caching, MOS is correlated with objective network metrics to predict user experience.

• Formula: Often estimated using the E-model (ITU-T G.107), which factors in latency and packet loss.
• Goal: A proactive caching system aims to maintain a MOS of > 4.0 for pre-fetched video and audio content.

A standard caching metric, but weighted by the QoS Class Identifier (QCI) of the request. A hit for a high-priority GBR flow is more valuable than a hit for a best-effort flow.

• Calculation: (Bytes served from cache for QCI 1-4) / (Total bytes requested for QCI 1-4).
• Optimization: The cache eviction policy is tuned to preferentially retain content associated with high-QoS slices.

A traffic shaping mechanism that controls the rate and burstiness of data transmission. Tokens are generated at a fixed rate and stored in a virtual bucket; a packet can only be transmitted if an equivalent number of tokens is available.

• Committed Information Rate (CIR): The guaranteed average rate at which tokens are replenished.
• Committed Burst Size (CBS): The maximum bucket capacity, defining the largest allowable traffic burst.
• Enforces a hard upper bound on bandwidth consumption for a specific cache service flow.

A traffic policing counterpart to the token bucket that smooths out bursty traffic by converting it into a steady, uniform stream. Packets are queued in a finite buffer and released at a constant rate, regardless of the ingress pattern.

• Smooths jitter: Ideal for real-time cached video delivery where consistent inter-packet spacing is critical.
• Enforces strict conformance: Non-conforming packets are discarded or marked as out-of-profile.
• Contrasts with the token bucket, which permits controlled bursts.

A scalable QoS architecture that classifies and manages network traffic by grouping flows into a limited number of service classes. Uses the DSCP (Differentiated Services Code Point) field in the IP header to mark packets.

• Per-Hop Behavior (PHB): Defines how each router treats a packet based on its DSCP marking.
• Expedited Forwarding (EF): Provides low-loss, low-latency, assured bandwidth—ideal for premium cached content.
• Assured Forwarding (AF): Offers different drop precedences within guaranteed rate classes.

A fine-grained QoS model that reserves dedicated end-to-end resources for each individual application flow using the RSVP (Resource Reservation Protocol) signaling protocol.

• Guaranteed Service: Provides firm mathematical bounds on delay and bandwidth for a specific flow.
• Controlled Load Service: Approximates the behavior of an unloaded network, even under congestion.
• Less scalable than DiffServ but offers absolute guarantees for mission-critical cache pre-fetching operations.

A set of techniques that proactively drop or mark packets before a router's queue becomes full, preventing bufferbloat and global TCP synchronization. Random Early Detection (RED) and CoDel are prominent algorithms.

• RED: Drops packets probabilistically based on average queue length, signaling congestion to TCP senders.
• CoDel (Controlled Delay): Targets packet sojourn time rather than queue length to control latency under load.
• Essential for maintaining low latency for cached content retrieval during traffic bursts.

A formal contract between a service provider and a customer that defines the measurable QoS guarantees for a delivered service. For cached content delivery, SLAs specify quantifiable performance targets.

• Key Metrics: Latency (e.g., 99th percentile < 50ms), throughput, jitter, and packet loss rate.
• Cache Hit Ratio SLA: Guarantees a minimum percentage of requests served from the edge cache.
• Penalties: Financial or service credit repercussions for failing to meet agreed-upon thresholds.