Opening the book…
Overview
- Chapter: 2 — Core Concepts of Distributed Systems
- Total Figures Proposed: 8
- Figure Types Used: Conceptual Illustration, Architecture Diagram, Flowchart, Comparison Table, Network Diagram, Infographic
Figure 2.1.1
Title: NIST Essential Characteristics of Cloud Computing
Detailed Caption: A navy header tile reading "Cloud Computing — NIST SP 800-145 Model" sits at the top, fanning out via dashed lines to a row of five teal characteristic tiles: On-Demand Self-Service (lightning icon), Broad Network Access (globe icon), Resource Pooling (rows icon), Rapid Elasticity (up/down arrows), and Measured Service (gauge), each with a one-line definition beneath the title. A second band labeled "Service Models" shows three light-gray panels for IaaS, PaaS, and SaaS with their full names and example components (Compute · Storage · Networking; Runtime · Middleware · Dev Tools; Applications · End-User Software). The bottom band, "Deployment Models," lists four warm-gold pills — Public, Private, Hybrid, Community — with short descriptions. The figure compresses the entire NIST definitional framework into a single one-page reference, so readers can locate any service offering against the official taxonomy at a glance.
Figure 2.2.1
Title: Evolution of Distributed Computing: From Clusters to Cloud
Detailed Caption: A horizontal navy timeline arrow carries four teal era circles spanning roughly six decades: Early Clusters & Mainframes (1960s–1980s, stacked-bars icon, "VAXcluster, HPC clusters, Tightly coupled"), Grid Computing (1990s–Early 2000s, networked-nodes icon, "SETI@home, Loosely coupled, Geographically distributed"), Cloud Computing (Mid-2000s, cloud icon, "AWS EC2/S3 (2006), Virtualization, Pay-as-you-go"), and Containers & Cloud-Native (2010s–Present, container icon, "Kubernetes, Microservices, Serverless/FaaS"). Beneath each circle, a "KEY TECH" callout lists representative technologies — shared memory and MPI for clusters, Condor and Globus Toolkit for grids, VMware and Xen for cloud, Docker and Service Mesh for cloud-native — and the right-most box is highlighted in a teal tint to mark the present. A bottom gradient bar runs from gray on the left to teal on the right, labeled "Less Abstraction → Maximum Abstraction." The figure positions today's containerized cloud not as a clean break from the past but as the latest stage in a long arc of increasing abstraction and scale, helping readers see where current tools inherited their ideas.
Figure 2.3.1
Title: Warehouse-Scale Computer: Datacenter as a Single Machine
Detailed Caption: Two side-by-side framed panels — "Single Computer" on the left (with a small desktop-monitor illustration) and "Warehouse-Scale Computer" on the right (with a row of server-rack icons) — bracket six horizontal pairs of components linked by teal "≡ analogous to" arrows. Reading top to bottom, the pairs map CPU/Cores ↔ Thousands of Servers, System Bus/Backplane ↔ Network Fabric, RAM ↔ Aggregate Memory (distributed), Hard Drive/SSD ↔ Distributed Storage (HDFS), Operating System ↔ Cluster Manager (Borg/K8s), and Power Supply & Fans ↔ Power & Cooling Infrastructure. Single-computer rows are shaded cream while WSC rows are shaded light teal, and a gold "ANALOGY" label sits between the two columns. The figure operationalizes Barroso and Hölzle's metaphor by showing the one-to-one mapping explicitly, so a student already familiar with PC internals can transfer that mental model directly to a hyperscale facility.
Figure 2.4.1
Title: Cluster Workload Scheduling and Resource Alignment
Detailed Caption: A navy "Cluster Manager / Scheduler" box at the top — annotated "Priority-based assignment · Over-subscription · Preemption" — fans arrows down to four worker-node columns labeled Node 1 through Node 4. Each node shows a stack of horizontal bars representing scheduled workloads: teal bars for high-priority services (guaranteed) and gold bars for low-priority batch jobs (preemptible), with a dashed red line marking node capacity. The four nodes intentionally illustrate four scheduling states — Node 1 is over-subscribed (bars exceed the capacity line), Node 2 shows preemption active (a faded gold bar being displaced), Node 3 is under capacity, and Node 4 is high-priority dense — and a legend below decodes the color/line semantics. By rendering the four states side by side, the figure makes Borg-style packing intuitive: the goal isn't to keep every node beneath the line, but to over-subscribe with preemptible work so that critical services still get their guarantees.
Figure 2.5.1
Title: Kubernetes Cluster Architecture
Detailed Caption: On the left, a large light-blue "Control Plane" panel encloses four components: a teal API Server ("Central communication hub") at the top, a teal Scheduler ("Assigns Pods to nodes") and Controller Manager ("Reconciles desired state") sitting beneath it with bidirectional arrows, and a gold etcd box ("Distributed key-value store · Cluster state") anchoring the panel as the single source of truth. To the right sit three Worker Node columns, each containing a teal kubelet at the top, a gray kube-proxy below it, and two or three light-teal Pod boxes labeled A through H, footed by "Container Runtime: containerd"; dashed arrows run from the API Server out to each kubelet to depict the API Server → kubelet command channel. A wide light-gray bar at the bottom — "Network Layer — Pod-to-Pod Communication across all Nodes" — emphasizes the flat overlay network. The diagram makes the architectural rule explicit: every component talks through the API Server, and etcd is the only place cluster state actually lives.
Figure 2.6.1
Title: HDFS Architecture: NameNode and DataNodes
Detailed Caption: A wide navy "NameNode (Master)" banner across the top — subtitled "Metadata & Namespace · File-to-block mapping · Replication policy" — fans dashed lines down to four DataNode columns labeled DataNode 1 through DataNode 4 (with a "...N DataNodes" ellipsis suggesting horizontal scaling). Each DataNode holds two or three colored block replicas — teal Block A, gold Block B, sage Block C — and the placement is engineered so that every block appears in exactly three different DataNodes, illustrated explicitly in a gold "3× Replication (default)" callout in the lower right (e.g., "Block A → DataNode 1, 2, 4"). A blue Client icon on the lower left shows the two-step access pattern with numbered arrows: ① "Lookup metadata" up to the NameNode, then ② "Read/Write data blocks" directly to a DataNode. The dashed lines back to the NameNode are labeled "Heartbeat & Block Report," reinforcing the HDFS lesson that metadata flows through the master while bulk data flows directly between client and DataNode — the key trick that lets a single master scale to petabytes.
Figure 2.7.1
Title: Netflix Microservices Architecture and Cloud Infrastructure
Detailed Caption: Three client devices on the far left — Mobile, Web, and Smart TV — funnel into a central navy "Zuul (API Gateway)" panel labeled "Routing & load balancing · Auth & rate limiting · Edge filtering," which then fans out to four teal microservice boxes stacked vertically: Auth Service, Catalog Service, Recommendation Service, and Playback Service, each with a two-line role description. Small "Ribbon LB" tags decorate the inter-service arrows to mark client-side load balancing, and a gold "Eureka (Service Registry)" panel on the right connects to every microservice with dashed bidirectional arrows for "Service discovery & registration · Health check & heartbeat." A separate light-red "Hystrix (Circuit Breaker)" callout — "Fault isolation · Fallback" — attaches via a dashed line to the Auth Service to indicate where circuit breakers wrap risky downstream calls. A wide bottom banner reads "AWS Cloud Infrastructure — EC2 · Auto Scaling · Multi-AZ · CloudFront CDN," anchoring the entire architecture in the cloud substrate; the figure as a whole shows how four named OSS components (Zuul, Eureka, Ribbon, Hystrix) cooperate to deliver the resilience patterns Netflix is famous for.
Figure 2.8.1
Title: Facebook's Layered Distributed Architecture
Detailed Caption: Four labeled tiers stack vertically — L1 Users / Clients (light gray, with user icons and the note "Web Browsers · iOS · Android · Desktop Apps — Billions of requests/day"), L2 Web Front-End Servers (navy, showing Server A, B, C plus "...N servers" for PHP/Hack application servers), L3 split between a wide teal Memcached Cluster ("Cache Node 1...N · Thousands of nodes") and a smaller "Specialized Stores" panel listing Haystack (photo/video) and Cassandra (NoSQL messages/feed), and L4 split between a gold Sharded MySQL ("Source of Truth," with Shards 1–3 partitioned by user range A–G, H–P, Q–Z, plus "...N shards, Primary-replica replication") and a parallel Hadoop / HDFS panel for analytics with HDFS and Hive/Spark sub-boxes. Directional arrows trace the request paths: clients down to front-ends, front-ends across labeled "Read path: cache first" into Memcached, Memcached down on a red "Cache miss → DB" arrow into MySQL, dashed "Writes → DB" lines into Specialized Stores, and a horizontal "ETL / Logs" arrow from MySQL into Hadoop. The figure replaces a multi-paragraph narrative with one diagram that shows where each Facebook subsystem lives, which paths are hot, and how the analytics pipeline is asynchronously fed off the source-of-truth shards.
This page titled 2.7.1: Chapter 2 was last modified on Tue, 22 Sep 2026 21:13:37 GMT and is shared under a CC BY 4.0 license and was authored, remixed, and/or curated by Felix Amoruwa.
The book could not be loaded just now.