A summary of the content of the cloud computing course CS3328

🧭 “Cloud Computing Technology” Knowledge Map (Final 2025 Edition)

1. The Meaning and Characteristics of the Cloud (Lecture 01)

• Definition of cloud computing: an on-demand, shared, metered service system of virtualized computing resources.

• Five essential characteristics (NIST):

  • On-Demand Self Service
  • Broad Network Access
  • Resource Pooling
  • Rapid Elasticity
  • Measured Service
• Three service layers: IaaS / PaaS / SaaS.
• Enabling technologies: virtualization, distributed computing, automation and SOA.

2. Cloud Models and Scenarios (Lecture 02)

• Deployment models: public cloud, private cloud, hybrid cloud, community cloud;
• Typical applications: government cloud, industrial cloud, AI training cloud;
• Evolution of service models: from resource hosting → platform orchestration → Function as a Service (FaaS);
• Industry trends: multi-cloud, edge-cloud, cloud-network convergence.

3. Distributed Computing in the Cloud (Lecture 03)

• MapReduce: the Map + Reduce (aggregation) model;
• Suitable scenarios: log analysis, text mining, sorting and statistics;

• The performance long-tail problem (Tail Latency):

  • caused by node heterogeneity, network fluctuations, etc.;
  • optimization strategies: task replication, progressive scheduling, asynchronous aggregation.

4. Warehouse-Scale Computers, WSC (Lecture 04)

• The WSC concept: an ultra-large-scale system that takes the data center as the unit of computation;
• Components: compute nodes, communication network, storage system, infrastructure;
• PUE (Power Usage Effectiveness): $$PUE = \frac{\text{total facility energy}}{\text{IT equipment energy}}$$
• Optimization directions: hot/cold aisle containment, liquid cooling, renewable energy.

5. Data Communication in the Cloud (Lecture 05)

• Evolution of storage systems: DAS → NAS → SAN → distributed file systems;
• Data access latency hierarchy: memory < SSD < HDD;

• Data center network topologies:

  • tree, Clos/Fat-Tree, Spine-Leaf;

• Communication optimization:

  • RDMA (Remote Direct Memory Access)
  • SDN (Software-Defined Networking) for programmable traffic.

6. Virtualization and Containers (Lecture 06)

• Virtualization levels: ISA (instruction set) → ABI (application binary) → API;

• Types of virtual machines:

  • System VM (KVM/Xen)
  • Process VM (JVM);
• Three principles (Popek-Goldberg): equivalence, resource control, efficiency;

• Development of containerization:

  • Docker and Kubernetes enable lightweight multi-tenancy;
  • the Cloud Native philosophy drives microservice architecture.

7. Workload Migration and Scheduling (Lecture 07)

• Scheduling hierarchy:

  • the guest OS schedules processes → vCPUs;
  • the hypervisor schedules vCPUs → pCPUs;

• Typical algorithms:

  • Xen Credit Scheduler: Weight + Cap;
  • supports preemption, fairness and proportional-share scheduling;

• Virtual machine live migration (Live Migration):

  • pre-copy and post-copy;
  • balancing downtime against migration latency.

8. Software-Defined Everything, SDDC (Lecture 08)

• Philosophy: compute, storage and networking are all virtualized and managed in a unified way;

• Typical architecture (VMware SDDC):

  • vSphere + NSX + Virtual SAN + vRealize;

• Resource pooling technologies:

  • Intel RSA, Google WSC;
• Key idea: Infrastructure as Code (IaC); software redraws the boundaries of hardware.

9. Cloud Computing Middleware (Lecture 09)

• Main types:

  • RPC (Remote Procedure Call)
  • MOM (message queues)
  • ORB (object request brokers)
  • data access middleware;

• Cloud-native orchestration systems:

  • Borg → Omega → Kubernetes;

• Scheduler architectures:

  • monolithic scheduler / two-level scheduling / shared-state scheduling;

• Role of middleware:

  • connects distributed services and improves scalability and fault tolerance.

10. Data Center Energy Saving (Lecture 10)

• Power model: $$P_d = A C V^2 f,\quad P_s = f(C)$$

• Energy-saving strategies:

  • DVFS (Dynamic Voltage and Frequency Scaling)
  • Race-to-Halt
  • EARtH (energy-aware scheduling)

• Energy proportionality (Energy-Proportionality):

  • server power is linearly related to load;
  • the goal is “the higher the load, the better the energy efficiency”.

11. Resource Utilization Optimization (Lecture 11)

• The resource wall problem:

  • the memory wall;
  • the dark silicon effect;

• The resource utilization dilemma:

  • background overhead, excessive redundancy;
  • co-scheduling of LC (Latency-Critical) vs BE (Best-Effort) tasks;

• Optimization directions:

  • Over-Provisioning and Over-Subscription;
  • peak shaving with UPS energy storage;
  • cross-layer power management: DVFS + task migration + energy modeling;
• Key goal: dynamically balance performance, reliability and energy consumption.

12. Reliability and Availability (Lecture 12)

(1) Reliability Models

• Failure Rate, mean time between failures (MTBF);
• the failure curve is “bathtub-shaped”: infant mortality → stable period → wear-out period;
• The reliability wall: the upper bound that fault-tolerance mechanisms place on parallel speedup.

(2) Levels of Redundancy

(3) Disaster Recovery and Backup

• Cold standby (Cold): low cost, high RTO;
• Warm standby (Warm): the DB is kept in sync;
• Hot standby (Hot): active-active data centers;
• Multi-Active: cross-region coordinated disaster recovery;
• The CAP principle: consistency, availability and partition tolerance cannot all be achieved at once.

(4) Security and Energy-Efficiency Attacks

• Multi-layer cloud security protection: tiered security at the IaaS/PaaS/SaaS layers;

• Security threats from a performance perspective:

  • malicious resource contention;
  • cache interference;
  • Power Attacks and Efficiency Attacks;

• Typical attack models:

  • manipulating power beyond the limit → UPS failure;
  • changing the access distribution → cache hit rate plummets.

🌐 Overall Knowledge Structure (Logical Main Line)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

graph TD
A[Cloud concepts and models] --> B[Distributed computing]
B --> C[Warehouse-scale computers]
C --> D[Data communication and storage]
D --> E[Virtualization and containers]
E --> F[Workload migration and scheduling]
F --> G[Software-defined data center]
G --> H[Middleware and resource scheduling]
H --> I[Data center energy saving]
I --> J[Resource utilization]
J --> K[Reliability and availability]

A --> A1[IaaS/PaaS/SaaS models]
B --> B1[MapReduce and tail latency]
C --> C1[PUE and energy efficiency]
D --> D1[SDN and distributed storage]
E --> E1[KVM/Docker/K8s virtualization layers]
F --> F1[Live migration and Xen scheduling]
G --> G1[SDDC and resource pooling]
H --> H1[Kubernetes and scheduling algorithms]
I --> I1[DVFS and race-to-halt]
J --> J1[LC/BE task scheduling and the power wall]
K --> K1[Redundancy and multi-site disaster recovery]
K --> K2[CAP theorem and security attacks]

Translated from the Chinese original.