Chapter 14: Network Design for Application Requirements

Application Profiling and Traffic Characterization

Application profiling is the systematic process of cataloging every application that traverses the network, documenting its traffic behavior, performance requirements, and business criticality. The output of profiling is a traffic matrix -- a table showing the volume and characteristics of traffic between every source-destination pair. This matrix drives link sizing, QoS policy design, and failover capacity planning.

Traffic characterization examines multiple dimensions of each application:

Latency, Jitter, and Loss Requirements

Different applications have dramatically different tolerances for network impairments. These thresholds are not arbitrary -- they derive from human perception and protocol behavior. The 150 ms one-way latency target for voice comes from ITU-T G.114.

The latency budget is your primary design constraint for real-time applications. A 150 ms one-way budget must account for serialization delay, propagation delay (~5 ms per 1,000 km of fiber), queuing delay, and codec processing delay.

QoS Design Framework

QoS is "managed unfairness, measured numerically in latency, jitter, and packet loss." The deployment framework follows seven steps: define business objectives, determine traffic classes, analyze application requirements, design platform-specific policies, test in controlled environments, pilot rollout, and production deployment with monitoring.

Design rules: Limit all Low-Latency Queues (LLQ) to 33% of aggregate link capacity. Reserve at least 25% for Best Effort traffic. Disable WRED on the LLQ; enable it on all Assured Forwarding classes.

Figure 14.1: QoS Deployment Framework -- seven-step process with continuous feedback

Trust Boundaries and Classification

Traffic should be classified and marked as close to the source as possible. The trust boundary defines where the network begins honoring endpoint markings:

• Conditionally trusted: Cisco IP phones, TelePresence endpoints, cameras (verified via CDP/LLDP)
• Trusted: Centrally managed PCs, managed wireless APs
• Untrusted: BYOD devices, printers, guest endpoints -- markings stripped and re-marked at ingress

At access-layer switches: policing on all edge ports (ingress), queuing on all switch ports (egress), minimum 1P3Q (one priority queue + three normal queues).

Application Dependency Mapping

ADM identifies relationships between applications, their supporting infrastructure, and communication patterns. It serves three design purposes: validation (ensures adequate connectivity for all dependency chains), risk identification (reveals hidden single points of failure), and migration planning (identifies which components must move together).

Figure 14.2: Application dependency map showing infrastructure relationships and potential single points of failure

Voice and Unified Communications

VoIP and UC place the strictest real-time requirements on the network. Core constraints: one-way delay ≤150 ms, jitter ≤30 ms (10 ms for TelePresence), packet loss ≤1% (0.05% for TelePresence), bandwidth 20-320 Kbps per call depending on codec.

WAN Design for Voice:

• LFI (Link Fragmentation and Interleaving): Essential on links below 768 Kbps -- fragments large packets and interleaves voice between fragments to reduce serialization delay
• QoS Preclassify: For VPN deployments, clones the original IP header before encryption so voice can be classified even when the payload is encrypted
• TCP Adjust-MSS: Rewrites SYN packets to prevent fragmentation with IPsec/GRE overhead (effective MTU drops to ~1,378 bytes)
• Call Admission Control (CAC): Limits simultaneous calls to match available bandwidth, rejecting new calls gracefully rather than degrading existing ones

Video Conferencing and Streaming Media

Video traffic falls into three distinct categories:

1. Interactive Video Conferencing: Real-time, 1-20 Mbps, latency ≤200 ms, jitter ≤50 ms. Mark CS4/AF41 with bandwidth guarantees.
2. Streaming Video (On-Demand): Buffered at client, tolerant of moderate jitter but needs sustained bandwidth. Mark AF31/32/33 with WRED.
3. Broadcast Video (Live IPTV): Uses multicast for scalability -- a 5 Mbps stream to 200 viewers consumes only 5 Mbps (vs. 1 Gbps with unicast).

Multicast Design Decisions:

IoT Network Design

IoT introduces challenges fundamentally different from traditional enterprise applications: massive device counts, resource-constrained hardware, diverse connectivity technologies, and critical security segmentation needs.

Security segmentation isolates IoT from critical business systems via VLANs, firewalls, or overlays for containment, policy enforcement, and visibility. Edge computing filters and aggregates data locally, reducing upstream bandwidth and enabling real-time analytics. MQTT publish-subscribe messaging scales better than client-server for thousands of devices.

Figure 14.3: IoT network architecture from device layer through edge computing to segmented core

Storage Replication and Backup Traffic

Storage traffic demands zero packet loss and often sustains high throughput for extended periods.

Data Center Bridging (DCB) enables lossless Ethernet for FCoE/RoCE: Priority Flow Control (PFC, 802.1Qbb) prevents frame loss per CoS; Enhanced Transmission Selection (ETS, 802.1Qaz) allocates bandwidth; DCBX auto-negotiates parameters between peers.

Replication modes: Synchronous replication (RPO=0, distance-limited to ~100 km due to speed-of-light latency on every write) vs. asynchronous replication (any distance, RPO of minutes to hours, batched writes).

Figure 14.4: Storage protocol and replication decision tree

Phased Implementation Strategies

The recommended implementation framework follows six steps:

1. Assessment: Evaluate users, devices, applications, and performance targets. Establish baselines.
2. Design Mapping: Create topology diagrams showing connections, backup paths, and traffic flows.
3. Security Integration: Layer firewalls, VLANs, IDS/IPS, and encryption into the design (not as an afterthought).
4. Installation and Configuration: Deploy with clear labeling, documented configs, static IPs for infrastructure.
5. Performance Testing: Stress-test under realistic load and optimize bottlenecks before production.
6. Monitoring and Maintenance: Continuous traffic observation, alerting, patching, lifecycle management.

Figure 14.5: Phased implementation framework with feedback loops

Migration Strategies

Four primary strategies with different risk-cost tradeoffs:

Figure 14.6: Migration strategy decision tree

Performance Baseline and Validation

Baselines established before migration serve as acceptance criteria afterward. Without a baseline, you cannot objectively determine whether the new design meets requirements.

Baseline metrics: end-to-end latency, jitter, packet loss, throughput utilization (peak, average, 95th percentile), application response times, error rates, failover times.

Validation sequence: pre-migration baseline, lab validation, pilot validation (compare to baseline), post-migration validation (acceptance thresholds), ongoing monitoring (detect drift).

A migration without a baseline is a leap of faith. A migration without a rollback plan is reckless. The CCDE exam expects designs that are measured, phased, and reversible.