Network Performance Optimization
Expert-defined terms from the Professional Certificate in Network Performance Testing Techniques course at London School of Business and Administration. Free to read, free to share, paired with a professional course.
Absolute Jitter – Concept #
The maximum deviation of packet inter‑arrival time from the expected interval. Related terms: jitter, latency, variation. Explanation: Absolute jitter measures the worst‑case timing error, useful for real‑time services where any spike can degrade quality. Example: A VoIP call experiences a 30 ms absolute jitter spike, causing audible distortion. Practical application: Network engineers monitor absolute jitter to set buffer sizes and QoS thresholds. Challenge: Isolating jitter sources in a multi‑hop path can be difficult without synchronized clocks.
Active Probing – Concept #
Technique that injects synthetic traffic to measure performance. Related terms: passive monitoring, synthetic transactions, measurement agents. Explanation: Tools generate test packets (e.g., ICMP, TCP SYN) and record round‑trip times, loss, and path characteristics. Example: Using iPerf to send UDP streams and capture throughput under load. Practical application: Active probing validates SLA compliance during maintenance windows. Challenge: Probing traffic may interfere with production traffic if not throttled appropriately.
Adaptive Congestion Control – Concept #
Algorithms that adjust sending rates based on real‑time network feedback. Related terms: TCP Cubic, BBR, flow control. Explanation: The sender monitors packet loss, RTT, and throughput, then dynamically scales the congestion window. Example: BBR estimates bottleneck bandwidth and RTT to maintain high utilization with low latency. Practical application: Data center fabric designers embed adaptive control to sustain performance under variable loads. Challenge: Mis‑tuned parameters can cause oscillations or unfair bandwidth sharing.
Aggregation Delay – Concept #
Time added when multiple packets are combined into a single frame for transmission. Related terms: frame aggregation, MTU, latency. Explanation: In wireless (e.g., 802.11n/ac) and Ethernet, aggregation improves efficiency but introduces a small queuing delay before the aggregated frame is sent. Example: A Wi‑Fi AP aggregates four 1500‑byte packets, adding ~2 ms delay. Practical application: Engineers balance aggregation size against latency‑sensitive traffic. Challenge: Determining optimal aggregation thresholds in mixed‑traffic environments.
Application‑Layer Performance – Concept #
End‑user perceived quality measured at the application level. Related terms: QoE, response time, throughput. Explanation: Metrics such as page load time, video start‑up delay, or database query latency reflect the combined effect of underlying network behavior. Example: A web app’s 2‑second load time corresponds to 150 ms network latency, 5 % packet loss, and server processing delay. Practical application: Performance testing tools (e.g., Selenium, JMeter) incorporate network throttling to simulate real conditions. Challenge: Isolating network impact from application code inefficiencies.
ARP Cache Poisoning – Concept #
Attack that corrupts the Address Resolution Protocol table to redirect traffic. Related terms: MITM, spoofing, security. Explanation: By sending forged ARP replies, an attacker forces devices to associate an IP address with the attacker’s MAC, intercepting or dropping packets. Example: A rogue host injects false ARP entries, causing all traffic to a server to flow through it. Practical application: Network performance monitoring includes ARP integrity checks to prevent hidden latency spikes. Challenge: Detecting subtle cache poisoning in large LANs without generating excessive alerts.
Asymmetric Routing – Concept #
Traffic takes different paths for forward and reverse directions. Related terms: path diversity, load balancing, routing loops. Explanation: When routes differ, measurement tools may report mismatched RTTs or loss, complicating performance analysis. Example: Packets from A→B travel via ISP‑1, while B→A travel via ISP‑2, resulting in 40 ms vs 70 ms RTT. Practical application: Engineers design symmetric routing policies for latency‑critical services. Challenge: Achieving symmetry in multi‑provider environments while maintaining redundancy.
Bandwidth Utilization – Concept #
Ratio of used bandwidth to available capacity. Related terms: throughput, link saturation, efficiency. Explanation: Expressed as a percentage, it indicates how well a link’s capacity is exploited. Example: A 10 Gbps link carrying 6 Gbps of traffic has 60 % utilization. Practical application: Capacity planning tools track utilization trends to predict upgrades. Challenge: High utilization can mask intermittent congestion, requiring granular time‑slice analysis.
Baseline Performance – Concept #
Reference measurements representing normal operating conditions. Related terms: historical data, benchmark, anomaly detection. Explanation: Establishing a baseline involves collecting latency, loss, jitter, and throughput over a period without disruptions. Example: A baseline shows average RTT of 12 ms with a standard deviation of 1 ms. Practical application: Alerting systems compare live metrics against baselines to spot deviations. Challenge: Baselines become stale after network changes, necessitating periodic re‑calibration.
Best‑Effort Traffic – Concept #
Traffic that receives no guaranteed delivery or latency service. Related terms: QoS, class of service, priority. Explanation: Packets are forwarded when resources are available; they may be delayed or dropped under congestion. Example: Standard web browsing traffic is typically best‑effort. Practical application: Network policies allocate guaranteed bandwidth to voice/video while relegating bulk transfers to best‑effort. Challenge: Excessive best‑effort load can still degrade other services if buffers fill.
Bidirectional Throughput – Concept #
Combined forward and reverse data rates measured simultaneously. Related terms: full‑duplex, symmetric bandwidth, traffic flow. Explanation: Some tests (e.g., TCP simultaneous open) report aggregate throughput, revealing duplex performance. Example: A 1 Gbps link shows 800 Mbps forward and 750 Mbps reverse, totaling 1.55 Gbps bidirectional. Practical application: Verifying that both directions meet service level expectations. Challenge: Asymmetric paths can cause one direction to dominate the measured total, obscuring bottlenecks.
Bufferbloat – Concept #
Excessive buffering causing high latency and jitter. Related terms: queue management, latency, congestion control. Explanation: Large buffers absorb bursts but increase queuing delay, especially noticeable in interactive applications. Example: A home router with a 1 MB queue adds 200 ms of latency under heavy download. Practical application: Deploying Active Queue Management (AQM) like CoDel reduces bufferbloat. Challenge: Balancing buffer size to prevent packet loss while maintaining low latency.
CARP (Cache‑Aware Routing Protocol) – Concept #
Routing protocol that considers content cache locations to optimize delivery. Related terms: CDN, edge caching, routing metrics. Explanation: CARP selects paths that traverse nodes with cached copies of requested objects, reducing origin fetch latency. Example: A video request follows a route through a POP that already stores the segment. Practical application: ISP networks implement CARP to improve streaming performance. Challenge: Maintaining cache state consistency and integrating with existing routing tables.
Carrier‑Grade NAT (CGN) – Concept #
Large‑scale NAT used by ISPs to conserve IPv4 addresses. Related terms: address translation, port exhaustion, latency. Explanation: CGN aggregates many subscriber sessions behind a few public IPs, introducing additional translation hops. Example: A residential user’s traffic passes through two NAT layers, adding 5 ms latency. Practical application: Performance testing includes CGN traversal to measure end‑to‑end delay. Challenge: Port‑overload can cause connection failures and unpredictable latency spikes.
Channel Utilization – Concept #
Percentage of a wireless channel’s capacity that is occupied. Related terms: spectral efficiency, airtime, interference. Explanation: High channel utilization indicates potential contention and increased contention windows. Example: A 20 MHz Wi‑Fi channel at 80 % utilization shows noticeable throughput drops for new clients. Practical application: RF engineers monitor utilization to adjust channel allocation. Challenge: Hidden nodes and external interference can cause utilization metrics to be misleading.
Churn Rate (Network) – Concept #
Frequency of topology changes such as link up/down events. Related terms: stability, convergence, routing updates. Explanation: High churn can trigger repeated route recomputation, temporarily degrading performance. Example: A dynamic MPLS network experiences 5 link flaps per hour, each causing a 30‑second convergence delay. Practical application: Monitoring churn helps plan redundancy and maintenance windows. Challenge: Distinguishing intentional reconfigurations from fault‑induced churn.
Circuit Switched Latency – Concept #
Delay incurred in circuit‑switched networks (e.g., PSTN) before data transmission. Related terms: setup time, call establishment, provisioning. Explanation: Unlike packet‑switched networks, a dedicated path is allocated, adding a fixed setup latency. Example: A voice call over a legacy circuit‑switched link incurs a 200 ms setup delay. Practical application: Comparing circuit‑switched latency with VoIP latency informs migration decisions. Challenge: Legacy systems may lack precise measurement hooks.
CoDel (Controlled Delay) – Concept #
AQP algorithm that aims to keep queue delay low. Related terms: bufferbloat, AQMs, latency reduction. Explanation: CoDel monitors minimum queue delay over intervals and drops packets when delay exceeds a target, forcing senders to reduce rates. Example: Enabling CoDel on a home router reduces ping spikes from 150 ms to under 20 ms. Practical application: Network devices embed CoDel to improve interactive traffic performance. Challenge: Tuning CoDel parameters for high‑speed links to avoid unnecessary drops.
Collision Domain – Concept #
Network segment where packets can collide, typically in Ethernet hubs. Related terms: broadcast domain, MAC, CSMA/CD. Explanation: Within a collision domain, simultaneous transmissions interfere, causing retransmissions and increased latency. Example: A 10 Mbps hub with four devices creates a single collision domain. Practical application: Segmenting networks with switches reduces collisions and improves performance. Challenge: Legacy infrastructure may still rely on hubs, limiting scalability.
Congestion Indicator (CI) – Concept #
Metric that signals the onset of congestion, often derived from ECN marks. Related terms: ECN, packet loss, RTT growth. Explanation: CI values are reported by routers to end‑hosts, prompting congestion‑avoidance behavior before loss occurs. Example: An ECN‑enabled path marks 0.2 % of packets, triggering TCP to halve its window. Practical application: Modern transport protocols (e.g., QUIC) use CI to maintain low latency. Challenge: Not all network equipment supports ECN, limiting visibility.
Connection Setup Time – Concept #
Duration from initial SYN to established TCP connection. Related terms: handshake latency, SYN, three‑way handshake. Explanation: Measured in milliseconds, it adds to overall transaction latency for short‑lived flows. Example: A web request experiences 45 ms connection setup due to remote server distance. Practical application: Persistent connections (keep‑alive) reduce repetitive setup overhead. Challenge: TCP SYN flooding attacks can artificially inflate setup times.
Content Delivery Network (CDN) – Concept #
Distributed system of edge servers that cache and serve content close to users. Related terms: edge caching, latency, PoP. Explanation: By reducing the hop count, CDNs lower RTT and improve throughput for large media files. Example: A video streamed from a CDN PoP 30 ms away versus 120 ms from the origin. Practical application: Performance engineers assess CDN placement to meet latency SLAs. Challenge: Cache miss penalties and dynamic content invalidation can introduce variability.
Cross‑Traffic Interference – Concept #
Impact of unrelated traffic flows on a target flow’s performance. Related terms: traffic shaping, QoS, contention. Explanation: Heavy background traffic can increase queueing delay, cause packet loss, and affect latency‑sensitive applications. Example: Bulk file transfers on the same VLAN cause VoIP jitter to exceed acceptable thresholds. Practical application: Deploying traffic classification and policing mitigates interference. Challenge: Accurately modeling cross‑traffic patterns in lab environments.
CTP (Congestion‑Tolerant Protocol) – Concept #
Transport protocol designed to operate efficiently under high congestion. Related terms: TCP, UDP, loss resilience. Explanation: CTP adapts its sending rate based on measured congestion signals while preserving data integrity. Example: In a satellite link with 500 ms RTT, CTP maintains steady throughput despite frequent loss bursts. Practical application: Remote sensing networks adopt CTP for reliable data delivery. Challenge: Limited deployment and compatibility with existing infrastructure.
Delay Variation (DV) – Concept #
Statistical measure of jitter, often expressed as standard deviation of packet delay. Related terms: jitter, latency, variance. Explanation: DV quantifies how much packet delay deviates from the mean, influencing real‑time application quality. Example: A VoIP stream with DV of 5 ms is considered excellent, while 30 ms degrades call clarity. Practical application: SLA contracts may specify maximum DV values. Challenge: Capturing accurate DV requires high‑resolution timestamping.
DiffServ (Differentiated Services) – Concept #
IP QoS architecture that classifies traffic into Per‑Hop Behaviors (PHBs). Related terms: DSCP, QoS, traffic prioritization. Explanation: Packets are marked with a DSCP value; routers apply forwarding treatment based on the PHB. Example: EF (Expedited Forwarding) DSCP 101110 ensures low‑latency handling for voice. Practical application: Enterprises configure edge routers to map application ports to DSCP values. Challenge: End‑to‑end DSCP preservation across multiple autonomous systems.
DNS Latency – Concept #
Time taken to resolve a domain name to an IP address. Related terms: resolver, caching, round‑trip time. Explanation: DNS queries add to total transaction latency, especially for first‑time lookups. Example: A DNS lookup of 85 ms contributes to a web page’s 250 ms load time. Practical application: Deploying local recursive resolvers reduces DNS latency. Challenge: DNS amplification attacks can overload resolvers, increasing response times.
Duplex Mismatch – Concept #
Inconsistent full‑duplex/half‑duplex settings between two Ethernet interfaces. Related terms: auto‑negotiation, collision, throughput. Explanation: One side operating full‑duplex while the other uses half‑duplex leads to collisions and severe throughput loss. Example: A 1 Gbps full‑duplex link paired with a half‑duplex NIC results in effective 10 Mbps throughput. Practical application: Network audits verify auto‑negotiation is enabled on all ports. Challenge: Legacy devices may only support fixed duplex modes.
ECN (Explicit Congestion Notification) – Concept #
IP header flag that signals congestion without dropping packets. Related terms: congestion indicator, TCP, QUIC. Explanation: When a router’s queue exceeds a threshold, it marks packets; receivers echo the mark back to the sender. Example: An ECN‑enabled path reduces packet loss from 0.5 % to 0 % while maintaining throughput. Practical application: Data centers enable ECN to achieve sub‑millisecond latency under high load. Challenge: Mixed‑vendor environments may have ECN disabled on some devices, limiting effectiveness.
Effective Bandwidth – Concept #
The usable portion of a link after accounting for protocol overhead and control traffic. Related terms: raw bandwidth, payload, efficiency. Explanation: For Ethernet, a 10 Gbps link may deliver ~9.5 Gbps of application data after headers and inter‑frame gaps. Example: A 100 Mbps Ethernet connection provides ~94 Mbps effective bandwidth for TCP payload. Practical application: Capacity planning tools factor effective bandwidth to avoid over‑provisioning. Challenge: Variable overhead from VPN encapsulation or tunneling can further reduce effective capacity.
ELI5 (Explain Like I’m 5) Performance Metric – Concept #
Simplified metric used to convey performance impact in non‑technical terms. Related terms: user experience, KPI, abstraction. Explanation: Translating technical latency figures into user‑perceived effects (e.g., “page loads in 2 seconds feels instant”). Example: A 100 ms increase in API latency is described as “noticeable delay” for end users. Practical application: Communicating performance improvements to business stakeholders. Challenge: Over‑simplification may hide underlying technical issues that need remediation.
End‑to‑End Latency – Concept #
Total time a packet takes from source application to destination application. Related terms: RTT, processing delay, propagation delay. Explanation: Sum of all network and processing components, measured in milliseconds. Example: A transaction from a web client to a backend service experiences 85 ms end‑to‑end latency. Practical application: SLA definitions often specify maximum end‑to‑end latency. Challenge: Isolating contributing factors requires detailed instrumentation at each hop.
Entitlement‑Based QoS – Concept #
QoS model where each user or service has a predefined bandwidth entitlement. Related terms: policing, shaping, service level. Explanation: Traffic exceeding entitlement is throttled or downgraded, ensuring fairness. Example: A tenant with a 200 Mbps entitlement cannot consume more than that on a shared fiber link. Practical application: Multi‑tenant data centers enforce entitlement QoS to avoid noisy neighbor effects. Challenge: Dynamic workloads may need flexible entitlement adjustments.
Ethernet Frame Size – Concept #
Total length of an Ethernet frame, including headers, payload, and CRC. Related terms: MTU, jumbo frames, overhead. Explanation: Standard frames are up to 1518 bytes; jumbo frames can be 9000 bytes, reducing per‑packet overhead. Example: Using 9000‑byte jumbo frames on a 10 Gbps link improves throughput by ~10 % for large file transfers. Practical application: Configuring consistent MTU across switches and NICs avoids fragmentation. Challenge: Some devices (e.g., older routers) do not support jumbo frames, leading to path MTU discovery failures.
Exponential Weighted Moving Average (EWMA) – Concept #
Smoothing algorithm used to estimate metrics like RTT. Related terms: RTT estimator, smoothing factor, variance. Explanation: EWMA combines the previous estimate with the latest measurement, weighted by a factor α (0 < α < 1). Example: TCP uses EWMA with α = 0.125 to compute Smoothed RTT (SRTT). Practical application: Accurate RTT estimates enable adaptive retransmission timers. Challenge: Selecting α that balances responsiveness against noise.
FEC (Forward Error Correction) – Concept #
Technique that adds redundant data to enable error recovery without retransmission. Related terms: packet loss, redundancy, coding. Explanation: Receivers can reconstruct lost packets using parity information. Example: In a satellite link, a Reed‑Solomon FEC scheme recovers up to 2 lost packets per 10‑packet block. Practical application: Real‑time video streaming over lossy networks employs FEC to maintain smooth playback. Challenge: Additional bandwidth overhead may reduce net throughput.
Flow Control – Concept #
Mechanism that prevents a sender from overwhelming a receiver’s buffer. Related terms: PAUSE frames, TCP window, backpressure. Explanation: In Ethernet, IEEE 802.3x PAUSE frames stop transmission temporarily; in TCP, the receiver advertises a window size. Example: A congested switch issues PAUSE frames to upstream devices, reducing inbound traffic. Practical application: Enabling flow control on NICs can improve lossless performance in storage networks. Challenge: Misconfigured flow control can cause head‑of‑line blocking across unrelated flows.
Four‑Tuple Flow – Concept #
Identification of a network flow using source IP, source port, destination IP, destination port. Related terms: flow table, NAT, session. Explanation: Devices like firewalls and load balancers track flows based on the four‑tuple to apply policies. Example: A TCP connection from 10.0.0.5:52314 to 192.168.1.10:443 defines one flow. Practical application: Performance monitoring tools aggregate statistics per four‑tuple to detect anomalies. Challenge: NAT devices rewrite source or destination addresses, complicating flow correlation.
FQ‑CoDel (Fair Queuing Controlled Delay) – Concept #
Queue management algorithm that combines fair queuing with CoDel’s latency control. Related terms: AQM, fairness, bufferbloat. Explanation: Each flow receives its own queue, and CoDel drops packets when its minimum delay exceeds a target, ensuring both fairness and low latency. Example: Linux kernel’s default qdisc FQ‑CoDel reduces ping spikes on congested home links. Practical application: Deploying FQ‑CoDel on edge routers improves interactive traffic performance. Challenge: Maintaining per‑flow state on high‑speed links can consume CPU resources.
Full‑Mesh Topology – Concept #
Network design where every node is directly connected to every other node. Related terms: redundancy, latency, scalability. Explanation: Provides minimal hop count and high resilience, but grows quadratically with node count. Example: A 10‑node full‑mesh fabric has 45 links. Practical application: Low‑latency clusters (e.g., HPC) use full‑mesh interconnects. Challenge: Cabling and port density become prohibitive beyond a moderate size.
GPRS (General Packet Radio Service) Latency – Concept #
End‑to‑end delay experienced over GPRS cellular networks. Related terms: cellular, RTT, network type. Explanation: Typically 300‑500 ms due to multiple radio and core network hops. Example: An IoT sensor using GPRS reports 420 ms latency to the cloud. Practical application: Designers account for high latency when selecting protocols (e.g., MQTT with QoS 0). Challenge: Variable signal strength can cause latency spikes and packet loss.
Granular QoS Policies – Concept #
Fine‑grained rules that apply QoS based on detailed attributes such as application, user, or device type. Related terms: policy‑based routing, class maps, ACLs. Explanation: Enables precise bandwidth allocation and priority handling. Example: Video conferencing traffic receives high priority, while bulk backup traffic is limited to 10 Mbps. Practical application: SD‑WAN orchestrators enforce granular QoS across branch sites. Challenge: Managing large numbers of policies can lead to configuration errors and increased processing overhead.
Ground‑Truth Measurement – Concept #
Baseline data collected from a controlled environment to validate testing tools. Related terms: benchmark, reference, validation. Explanation: Provides an accurate reference point against which field measurements are compared. Example: Lab testing of a new router shows 0.5 ms latency under load; field measurements deviating by >10 % trigger investigation. Practical application: Certification labs use ground‑truth to certify equipment performance. Challenge: Replicating real‑world traffic patterns in a lab is often impractical.
Handoff Latency – Concept #
Delay incurred when a mobile device switches between access points or cells. Related terms: roaming, handover, interruption. Explanation: Includes authentication, re‑association, and path re‑optimization. Example: Wi‑Fi handoff latency of 70 ms can cause brief video freeze. Practical application: Optimizing AP placement and using fast‑roaming standards (e.g., 802.11r) reduces handoff latency. Challenge: Heterogeneous networks (Wi‑Fi + LTE) have differing handoff mechanisms, complicating unified measurement.
HEP (High‑Efficiency Packet) Header – Concept #
Streamlined packet header designed for low‑overhead transmission in high‑throughput environments. Related terms: encapsulation, protocol overhead, latency. Explanation: Reduces header size from typical 40 bytes (IPv4 + TCP) to as low as 12 bytes, improving payload efficiency. Example: In a data‑center fabric, HEP reduces per‑packet latency by 0.2 µs. Practical application: Custom ASICs in hyperscale data centers adopt HEP for latency‑critical services. Challenge: Interoperability with standard IP stacks requires translation gateways.
Header Compression – Concept #
Technique that reduces the size of packet headers on links with limited bandwidth. Related terms: ROHC, IPHC, overhead reduction. Explanation: By sending only changes between successive headers, bandwidth is saved and latency reduced. Example: Over a 64 kbps satellite link, ROHC compresses IPv4/TCP headers from 40 bytes to 3 bytes. Practical application: Mobile backhaul links use header compression to improve throughput. Challenge: Lossy links can cause desynchronization, requiring periodic full‑header refreshes.
Hybrid WAN Optimization – Concept #
Combination of traffic shaping, caching, and protocol acceleration across mixed‑technology WANs. Related terms: SD‑WAN, WAN accelerator, deduplication. Explanation: Optimizes performance by applying different techniques based on path characteristics. Example: A branch office uses compression over MPLS and deduplication over broadband simultaneously. Practical application: Enterprises deploy hybrid solutions to meet latency targets without full MPLS migration. Challenge: Managing policy conflicts between overlapping optimization layers.
ICMP Rate Limiting – Concept #
Control mechanism that restricts the number of ICMP messages a device can generate per second. Related terms: ping flood, security, measurement impact. Explanation: Prevents denial‑of‑service attacks that exploit ICMP echo requests but may affect active probing accuracy. Example: A router limited to 10 ICMP replies per second may cause ping‑based latency tests to report intermittent loss. Practical application: Configuring appropriate rate limits balances security and monitoring needs. Challenge: Selecting limits that are low enough to mitigate attacks yet high enough for legitimate diagnostics.
In‑band OAM (Operations, Administration, and Maintenance) – Concept #
OAM functions that operate within the data traffic stream. Related terms: Y.1731, BFD, performance monitoring. Explanation: Devices embed OAM packets (e.g., continuity check messages) alongside user data, enabling real‑time fault detection. Example: Ethernet OAM frames verify link health without dedicated out‑of‑band channels. Practical application: Service providers use in‑band OAM to monitor SLA metrics. Challenge: OAM traffic can be deprioritized by misconfigured QoS, reducing visibility.
Ingress Filtering – Concept #
Security practice that discards packets with spoofed source addresses at network entry points. Related terms: anti‑spoofing, BCP 38, packet loss. Explanation: Helps prevent DDoS amplification and reduces unnecessary traffic that could affect performance. Example: An ISP implements ingress filtering, dropping packets claiming to originate from private IP ranges. Practical application: Reducing unwanted traffic improves overall latency and bandwidth availability. Challenge: Incorrect filter rules can inadvertently block legitimate traffic, causing performance degradation.
Inter‑Packet Gap (IPG) – Concept #
Mandatory idle period between Ethernet frames to allow receivers to process bits. Related terms: ethernet timing, overhead, serialization. Explanation: Standard IPG is 96 bit times (≈9.6 µs at 10 Mbps). Example: On a 1 Gbps link, the IPG adds ~0.96 µs per frame, negligible for large frames but noticeable for many small packets. Practical application: Timing analysis for high‑frequency trading systems accounts for IPG latency. Challenge: Non‑standard IPG settings can cause compatibility issues with other devices.
IPsec Overhead – Concept #
Additional bytes added by IPsec encapsulation (ESP/AH) to secure traffic. Related terms: VPN, encryption, MTU. Explanation: ESP adds 24 bytes (without authentication) or up to 56 bytes with authentication and padding. Example: A 1500‑byte Ethernet frame carrying IPsec ESP may need fragmentation to fit the path MTU. Practical application: Network planners account for IPsec overhead when sizing WAN links. Challenge: Excessive overhead reduces effective bandwidth and may increase latency due to additional processing.
Jitter Buffer – Concept #
Memory used by receivers to smooth out packet arrival time variations. Related terms: jitter, latency, playout delay. Explanation: Packets are stored briefly; late packets are discarded, early packets are delayed to maintain continuous playback. Example: A VoIP handset with a 30 ms jitter buffer masks modest network jitter. Practical application: Adjusting jitter buffer size balances latency against packet loss. Challenge: Too large a buffer adds perceptible delay; too small leads to choppy audio.
Keep‑Alive Mechanism – Concept #
Periodic messages sent to maintain a session’s active state. Related terms: heartbeat, session timeout, TCP keep‑alive. Explanation: Prevents idle connections from being closed by intermediate devices. Example: An HTTP/2 connection uses PING frames every 30 seconds to keep the tunnel alive. Practical application: Ensuring persistent connections for low‑latency APIs. Challenge: Excessive keep‑alive traffic can add unnecessary load on constrained networks.
Latency Budget – Concept #
Allocation of allowable delay across each segment of a network path to meet end‑to‑end targets. Related terms: budgeting, SLA, timing analysis. Explanation: Engineers assign portions of total latency to propagation, queuing, processing, and serialization. Example: For a 100 ms end‑to‑end SLA, 20 ms is reserved for propagation, 30 ms for queuing, 10 ms for processing, leaving 40 ms margin. Practical application: Designing data‑center networks with strict latency budgets for financial trading. Challenge: Unexpected congestion can consume budgeted queuing time, violating SLA.
Link Aggregation (LAG) – Concept #
Combining multiple physical links into a single logical interface for higher bandwidth and redundancy. Related terms: IEEE 802.3ad, LACP, port‑channel. Explanation: Traffic is load‑balanced across member links using hash algorithms. Example: Four 10 Gbps ports aggregated to provide a 40 Gbps LAG. Practical application: Core switches use LAG to increase capacity while providing failover. Challenge: Hash collisions may cause uneven utilization, and mis‑aligned LACP settings can cause link flapping.
Load‑Balancing Algorithms – Concept #
Methods used to distribute traffic across multiple paths or links. Related terms: hash, round‑robin, ECMP. Explanation: Algorithms consider fields such as source/destination IP, ports, or flow identifiers to achieve fairness. Example: ECMP (Equal‑Cost Multi‑Path) splits traffic across equal‑cost routes using a 5‑tuple hash. Practical application: Data‑center fabrics rely on ECMP to scale bandwidth. Challenge: Changing traffic patterns can cause hash imbalance, leading to hot links.
Loss Concealment – Concept #
Techniques that mask packet loss in real‑time media streams. Related terms: PLC, error concealment, jitter buffer. Explanation: Receivers generate synthetic audio or video frames to fill gaps. Example: A VoIP receiver synthesizes a 20 ms audio segment when a packet is missing. Practical application: Maintaining call quality over lossy networks. Challenge: Excessive loss can overwhelm concealment algorithms, resulting in audible artifacts.
MAC Learning – Concept #
Process by which a switch builds its forwarding table based on source MAC addresses of received frames. Related terms: CAM table, flooding, aging. Explanation: When a switch sees a frame, it records the MAC and associated port; subsequent frames to that MAC are forwarded directly. Example: A new host sends a frame; the switch learns its MAC and updates the table. Practical application: Accurate MAC learning reduces unnecessary flooding, improving efficiency. Challenge: MAC address spoofing can corrupt tables, leading to security and performance issues.
Maximum Transmission Unit (MTU) – Concept #
Largest packet size that can be transmitted without fragmentation. Related terms: jumbo frames, path MTU, fragmentation. Explanation: MTU is defined per interface; mismatched MTUs cause Path MTU Discovery (PMTUD) to trigger. Example: A 1500‑byte Ethernet MTU vs. a 9000‑byte jumbo frame on a data‑center link. Practical application: Setting consistent MTU across a network avoids packet loss due to oversized frames. Challenge: Some middleboxes drop ICMP “fragmentation needed” messages, breaking PMTUD.
Mean Opinion Score (MOS) – Concept #
Subjective quality rating for voice or video, ranging from 1 (bad) to 5 (excellent). Related terms: QoE, subjective testing, codec. Explanation: MOS correlates with objective metrics such as packet loss, jitter, and latency. Example: A VoIP call with 0.5 % loss and 30 ms jitter yields MOS ≈ 4.2. Practical application: Service providers report MOS values to illustrate perceived quality. Challenge: MOS does not capture all user expectations, especially for interactive applications.
Microburst Traffic – Concept #
Sudden, short‑duration spikes of high‑rate traffic that can overflow buffers. Related terms: bufferbloat, burstiness, queuing delay. Explanation: Even brief bursts can cause packet loss if buffers are insufficient. Example: A server sends a 1 GB file in a burst lasting 10 ms, exceeding a 500 Mbps link’s instantaneous capacity. Practical application: Traffic shaping smooths bursts to protect latency‑sensitive flows. Challenge: Detecting microbursts requires high‑resolution monitoring.
Multipath TCP (MPTCP) – Concept #
Extension of TCP that allows a single connection to use multiple paths simultaneously. Related terms: subflow, congestion control, redundancy. Explanation: MPTCP splits data across subflows, each with its own congestion window, improving throughput and resilience. Example: A mobile device combines Wi‑Fi and LTE links for a single MPTCP session, achieving aggregate bandwidth. Practical application: Data‑center servers use MPTCP to balance load across multiple NICs. Challenge: Middleboxes that do not understand MPTCP may drop subflow packets.
NetFlow Sampling – Concept #
Technique that records a subset of packets to generate traffic statistics. Related terms: sFlow, IPFIX, flow export. Explanation: Sampling ratios (e.g., 1 out of 1000) reduce processing overhead while providing representative data. Example: A router samples 1 % of packets for NetFlow, exporting flow records to a collector. Practical application: Capacity planning and anomaly detection rely on sampled flow data. Challenge: Low‑volume traffic may be missed, leading to inaccurate metrics.
Network Coding – Concept #
Method of mixing multiple data streams at intermediate nodes to improve throughput and robustness. Related terms: linear coding, multicast, resilience. Explanation: Receivers decode original packets using linear combinations, allowing recovery from losses without retransmission. Example: In a wireless mesh, network coding reduces required retransmissions by 30 %. Practical application: Satellite constellations use coding to mitigate high error rates. Challenge: Increased computational complexity and the need for synchronized decoding.
Network Function Virtualization #
Network Function Virtualization