• Network latency is the time delay between when data is sent from source to destination - measured as round-trip time (RTT) or one-way delay
• Caused by multiple factors that accumulate as packets traverse the network path
• Critical performance metric that affects user experience, especially for real-time applications (VoIP, video conferencing, gaming)
• Different from bandwidth (capacity) - you can have high bandwidth but still experience high latency

Types of Latency

• Propagation Delay: Time for signal to travel through transmission medium
  • Fiber optic: ~5 microseconds per kilometer
  • Copper: ~5.5 microseconds per kilometer
  • Satellite: 250-280ms (geostationary orbit)
• Transmission Delay: Time to push all packet bits onto the wire
  • Formula: Packet size / Link bandwidth
  • Example: 1500-byte frame on FastEthernet = 1500×8 / 100Mbps = 0.12ms
• Processing Delay: Time for device to examine packet headers and make forwarding decisions
  • Routers: 1-10ms (varies by hardware/software)
  • Switches: <1ms (hardware-based forwarding)
• Queuing Delay: Time packet waits in output queue before transmission
  • Variable based on network congestion and QoS policies

Common Latency Sources

• WAN Links: Especially satellite (250-500ms RTT) and long-distance terrestrial circuits
• Router Hops: Each Layer 3 hop adds processing delay
  • Rule of thumb: 1-5ms per router hop on modern equipment
• Network Congestion: Causes increased queuing delays and potential packet loss
• Protocol Overhead: TCP acknowledgments, retransmissions, and window scaling
• DNS Resolution: Can add 20-100ms before actual data transfer begins

Measuring Latency

Latency Requirements by Application

Latency Optimization Techniques

• QoS Implementation: Prioritize time-sensitive traffic using DSCP markings
  • Voice: EF (Expedited Forwarding) - 46
  • Video: AF41 (Assured Forwarding) - 34
• Traffic Shaping: Prevent buffer bloat by controlling burst rates
• Path Optimization: Use routing protocols with latency-aware metrics
  • EIGRP: Includes delay in composite metric calculation
  • OSPF: Can use delay-based cost modifications
• Local Caching: Content Delivery Networks (CDNs) and proxy caches reduce RTT
• Protocol Tuning: TCP window scaling, selective acknowledgments (SACK)

Vocabulary

• RTT (Round-Trip Time): Total time for packet to travel to destination and acknowledgment to return
• Jitter: Variation in latency over time - critical for real-time applications
• Buffer Bloat: Excessive buffering causing increased latency under load
• Cut-Through Switching: Forwards frames before complete reception (reduces latency vs store-and-forward)
• Serialization Delay: Time to convert digital data to signals on physical medium

Notes

• Latency is often more noticeable to users than bandwidth limitations - 10Mbps with 20ms latency feels faster than 100Mbps with 200ms latency for interactive applications
• Satellite connections have inherent high latency due to distance (22,236 miles to geostationary orbit) - no amount of bandwidth can overcome physics
• Use ping -t (Windows) or ping without count (Linux) for continuous latency monitoring during troubleshooting
• Consider asymmetric latency - upload and download paths may have different delays, especially with satellite or cellular connections
• QoS policies must be implemented end-to-end to be effective - single bottleneck point can negate optimization efforts elsewhere
• Modern applications use techniques like prefetching and connection pooling to mask latency effects from users