SYSTEMS & ENGINEERING · VIRTUAL-FIRST REFERENCE
Home Connectivity, Computing, Cybersecurity & AI
A layered reference for building or simulating technical environments at home: obtain connectivity, understand equipment, model systems virtually, protect information, then add compute and AI workflows. Not a statement that these environments have been purchased, installed or audited.
01 · THE CONNECTION
Where the internet comes from
The global internet is a network of independently operated networks exchanging traffic. A household normally receives internet access through an internet service provider (ISP). Fibre, cable, DSL, cellular and satellite are different access paths. A router creates a local network; running one does not create an independent internet uplink.
Internet access
ISP service, permitted mobile tethering or other contracted uplinks. Alternative routes differ in coverage, speed, latency and total cost.
Local-only network
Devices can share data, printers and local services through a router or switch without an external internet connection. Useful for isolated experiments.
Hosting and publishing
A locally hosted service still needs suitable connectivity, DNS, secure external access and service maintenance. A static website can instead be hosted on GitHub Pages.
02 · PHYSICAL ARCHITECTURE
What the equipment does
| Component | Purpose | Physical alternative / decision |
|---|---|---|
| ONT / modem | Terminates the provider's incoming service | Often supplied by ISP; cannot be meaningfully replaced by a simulated modem for real service |
| Router / NAT | Connects networks and forwards traffic | ISP router, mini PC router or router VM in a test environment |
| Firewall | Applies defined traffic policy | Integrated router firewall, host firewall or isolated virtual firewall |
| Switch / AP | Wired network expansion / wireless access | Integrated router ports and Wi-Fi first; add managed switch or AP only if needed |
| Compute host | Runs OSs, VMs and AI software | Existing laptop, desktop, mini PC or hosted computing service |
| Storage / backup | Preserves independent copies and restoration points | Encrypted external drive, managed backup or NAS |
| GPU / accelerator | Speeds selected graphics and parallel compute tasks | Integrated GPU, discrete GPU or hosted accelerated runtime |
| Power / cooling | Keeps equipment within safe operating conditions | Ventilated shelf first; UPS and rack only when requirements justify them |
A server rack is an enclosure, not a prerequisite for networking or AI. Rack illustrations are useful for identifying components, but do not imply that every home requires a rack, high-voltage equipment or complex cabling.
03 · VIRTUAL BEFORE PURCHASING
Free and lower-cost ways to model the system
Three types of substitute must be distinguished: a simulator represents network behaviour; a virtual machine runs operating-system software on existing hardware; and a hosted service supplies remote compute or storage. None automatically replaces a household's physical ISP connection.
Cisco Packet Tracer
Network topology and configuration practice. Education tool, free with eligible registration; not a substitute for a real ISP or production firewall.
GNS3
Free/open-source network modelling with virtual devices. Some vendor images are separately licensed. Hardware resources determine scale.
OPNsense
Runs as a virtual firewall on supported hypervisors. Best introduced within an isolated lab before considering any change to a functioning household router.
VirtualBox
Virtual machines for isolated operating-system and software experiments. Depends on the host's CPU, RAM and storage.
Google Colab
Notebook-based experiments with variable free resources and restrictions. Accelerator access is not guaranteed.
Ollama
A local model runner for supported machines. Model suitability depends on memory, model format, context and system workload.
Suggested virtual-first concept sequence
First draw the topology and check the devices you already own. Next model routing and segmentation in a simulator. Then examine a virtual firewall in an isolated VM. Finally compare local versus hosted AI compute. These are staged learning objectives, not instructions to alter an existing household network.
03A · CLOUD COMPUTING & VIRTUAL NETWORKS
Cloud machines, serverless services and a virtual network
The additional examples show a virtual machine running on a provider's infrastructure, a network of isolated subnets, hosted GPU options and serverless data processing. These are not ways to generate an internet connection: they still require internet access from a home or other uplink to administer them remotely.
Azure VM / hosted operating system
A cloud VM has its own operating system, virtual CPU, memory, disk and network configuration. Public access requires deliberate security decisions; a running VM can incur charges. Microsoft's official Azure portal quickstart.
Subnets, routes and security groups
A virtual network defines address ranges and connectivity policies. A public subnet, private subnet and security group serve different purposes. A virtual network is not automatically publicly reachable or free.
Functions, queues, events and storage
Managed functions and stream/event services run without you operating a persistent VM. Charges can still arise from invocations, storage, traffic and logs. The uploaded comparison image contains older vendor and product names and should not be copied as a current catalogue.
GPU or virtual workstation
Hosted accelerators are useful for occasional workloads but are subject to quota, price, availability, data handling requirements and software licensing. No guaranteed free GPU is implied.
Cloud-provider alternatives and limits
| Option | Practical role | Free status / restriction |
|---|---|---|
| Google Cloud Free Program | Small hosted VM, limited storage and other selected services | Eligible regional e2-micro allocation; GPUs and TPUs excluded from the Always Free offering |
| Microsoft Azure | Virtual machines, VNet, cloud disks and managed services | Account/trial eligibility and current billing controls must be checked; VM creation does not mean free use |
| GitHub Codespaces | Cloud development workspace for coding and testing | Personal-account included compute and storage quotas; not a free permanent server or GPU entitlement |
| Google Colab | Notebook exploration and occasional acceleration | Free access, accelerator availability and runtime duration vary; not a reserved GPU service |
| VirtualBox on existing computer | Isolated simulated computing environment | Can avoid cloud charges, but consumes local CPU/RAM/storage and still relies on existing hardware |
Priority: model and explain systems on equipment already available, use time-limited hosted resources only where necessary, and keep cloud billing limits and data privacy explicit. An image claiming free daily model hours or a sign-up bonus is not verification of provider identity, eligibility or continued availability.
03C · NETWORK TOPOLOGIES & CONNECTIVITY CHOICES
From a single room to a multi-site network
These references distinguish equipment, transmission links and policy. A physical topology shows what connects to what; a logical topology shows addressing, VLAN boundaries, routing and access decisions. Network diagrams are models, not proof of deployed infrastructure.
Single router, wired and hybrid
For a small coverage area, start with one well-positioned router. A switch adds Ethernet ports, not a new internet subscription. Use wired access points, mesh nodes or a mixed arrangement if a measured coverage problem justifies them.
Extender, mesh, access point and Wi‑Fi 7
Repeaters may reduce effective throughput when sharing a radio for backhaul; not all repeaters halve it and not every extender creates a second network name. Mesh nodes vary by backhaul and placement. Wi‑Fi 7 multi-link operation (MLO) is a capability of compatible access points and clients; it cannot guarantee zero jitter or increased internet speeds.
VLANs and inter-VLAN policies
An access port normally serves one VLAN; 802.1Q trunks can carry multiple tagged VLANs. A router or Layer 3 switch can route between them, but security requires explicit traffic rules. Guest and connected-device separation should be validated, not assumed from VLAN labels alone.
Wi‑Fi, wired, Zigbee and monitoring
Ethernet generally gives predictable performance for fixed equipment. Wi‑Fi provides flexible wireless IP connectivity. Zigbee is a low-power device-network protocol often using a coordinator or hub; the comparison is about device needs, not one universal winner. Solar monitoring controllers may use cellular, Ethernet or Wi‑Fi uplinks.
WAN, VPN, site-to-site and telephony
A wide-area network connects geographically separate sites. Virtual private network (VPN) tunnels can connect authorized users or networks over an existing carrier link. Traditional PSTN gateways and contemporary internet-based voice services are separate components; older architecture diagrams may reflect legacy telephony deployments.
Central hub, VPN, Interconnect and SD‑WAN
A cloud connectivity hub can link VPC networks and external sites. Dedicated or partner interconnect, VPN and SD‑WAN represent distinct connectivity arrangements, with different commercial and operational requirements. None constitutes free household internet access.
Original comparison: what each example helps explain
| Diagram class | Question illustrated | Important qualification |
|---|---|---|
| ISP → ONT / modem → router → devices | Where does the connection originate? | Many provider gateways combine modem, router and access point in one box |
| Router → switch → NAS / PC / AP | How is one local connection distributed? | A switch does not independently supply internet access; not every device has Ethernet |
| VLAN 10 / 20 / 30 | How can several logical networks share equipment? | Isolation depends on routing and firewall policy, not illustration colours |
| Wi‑Fi extender / mesh / wired AP | How can weak coverage be improved? | Floor area alone is not a reliable selection criterion; measure walls, interference and backhaul |
| WAN / branch office / VPN | How can distant offices interoperate? | External services and authenticated access are additional requirements |
| Cloud VPC hub with hybrid spokes | How can cloud and non-cloud networks communicate? | Provider-specific functions, routing and charges must be verified |
Sources and further reading: Cisco on 802.1Q and inter-VLAN routing; Google Cloud Network Connectivity Center; Wi‑Fi 7 MLO compatibility and requirements. Brand/vendor illustrations provided for comparison are not reproduced or endorsed.
03B · OBSERVATION & NETWORK DESIGN
Wi‑Fi, VLANs, telemetry and switch architecture
The new imagery connects the back of a home router to bigger subjects: subnet boundaries, virtual networks, VLAN separation, link aggregation and performance monitoring. A management dashboard can report CPU, network, GPU and memory utilisation, but readings must be interpreted against the device and task measured.
Wi‑Fi router hardware
Compare WAN, LAN, Ethernet ports, radio bands, antenna design, firmware support and wireless coverage. A device labelled Wi‑Fi 6E must support the relevant bands; the router does not inherently increase the ISP subscription speed.
VLANs and switching
Logical traffic separation needs compatible switching/routing and an enforceable policy. Vendor-specific switch command illustrations are not universal. PortFast should be understood only in the correct edge-port and spanning-tree context.
Link aggregation and vPC
Bundled links may provide redundancy and aggregate capacity, not necessarily a faster single flow. Cisco vPC is a distinct two-switch architecture with platform-specific requirements, unsuitable as a default home requirement.
Performance metrics
Measure latency, jitter, packet loss, throughput, device load, and recovery times separately. An appealing dashboard is not evidence of improved performance until tests are recorded.
05A · CHOOSING ACCELERATION RESPONSIBLY
GPU optimisation and hosted resources
Start with supported drivers, OS updates, ordinary device temperature monitoring, actual workload measurements and vendor-documented settings. Changes to firmware, voltage or clock speed can create instability and should not be treated as default optimisation steps. Cloud GPU marketing claims about training speed or certification require independent verification.
Distinguish a graphics card, a GPU-equipped workstation, a virtual GPU allocation, a cloud GPU instance and a shared hosted AI service. They differ in ownership, cost, isolation, software control and available resources.
04 · CYBERSECURITY
Security controls in the right places
Firewalls
Network controls may operate at packet, connection and application levels. A web application firewall protects HTTP-oriented applications; it does not replace endpoint protection.
Endpoint security
Keep operating systems and applications updated; use device encryption, appropriate anti-malware protection and supported security tools. EDR provides additional monitoring where justified.
Identity
Unique credentials, a password manager, multi-factor authentication and carefully scoped administrative permissions protect accounts and services.
Segmentation
Keep guest or smart-home devices separate from sensitive research devices where supported by the router. Network separation must be verified, not inferred from a label.
Backups
Use independent, versioned, protected copies and periodically verify restoration. Backup and system availability are different requirements.
Monitoring
Maintain relevant security logs, alerts and response responsibilities. AI-assisted classification can help prioritise review but is not a guarantee of detection.
How the security diagrams should be interpreted
The OSI-layer diagrams are educational approximations, not an exhaustive classification of products; protective functions overlap layers. DDoS, DNS spoofing, person-in-the-middle activity, botnets and malware illustrate distinct risks. One supplied graphic labels an ARP-related example as IP spoofing; those concepts should not be treated as interchangeable. The home firewall rules image is a schematic example, not a universal configuration to copy.
Related workstream: cybersecurity and backup explanation.
05 · COMPUTE / ACCELERATION
When CPUs, GPUs and remote accelerators matter
| Option | Suitable for | Limitation |
|---|---|---|
| Existing CPU / integrated GPU | Network labs, coding, admin, lightweight inference | Slower for some large parallel workloads |
| Local discrete GPU | Graphics, supported AI inference and training experiments | VRAM, heat, electricity and cost constraints |
| Cloud GPU / notebook | Occasional acceleration without owning a powerful card | Time limits, quotas, network reliance, data handling and fees |
| Multi-GPU / interconnect | Specialised large-model workloads | Complexity, compatibility, hardware cost and communication overhead |
| Hosted AI API | AI features without managing model-serving hardware | Provider terms, ongoing cost, privacy and model availability |
The uploaded VRAM/model matrices and NVLink numbers are useful comparison prompts, not verified compatibility guarantees. Model size, quantisation, context length, KV cache, software compatibility and available system memory all affect practical performance.
06 · INTEGRATE AI AND AUTOMATION
A controlled workflow from data to output
Small workflows can support news aggregation, correspondence drafts, source-to-claim verification, backup alerts, document indexes and research evaluation. Sensitive information and publication actions should remain behind permission and review controls.
NON-EXECUTABLE ARCHITECTURE ANNOTATION
SOURCE → validation → processing → evidence check
→ human approval → output
→ timestamped version history
News synthesis · Email approval workflows · AI assurance · Business automation
07 · ALTERNATIVE COMBINATIONS
Setups compared by purpose, not expense
Existing equipment
ISP router + current laptop or mobile device + appropriately secured backups and supported cloud services.
Virtual learning lab
Existing computer + network simulator + isolated virtual machines; no rack or GPU purchase required.
Research and automation
Source registry + protected data storage + notebooks or hosted APIs + reviewed outputs + evidence log.
Local AI experimentation
Device with suitable memory + local model runtime + benchmark and safety controls, with hosted acceleration only where needed.
Extended home infrastructure
Switches, APs, storage and power protection added only to satisfy evidenced reliability, capacity or recovery requirements.
08 · ORIGINAL REFERENCES
Source documentation and image coverage
The illustrative diagrams on this page are independently authored in HTML. Third-party screenshots and infographic designs supplied for reference are not republished as website assets or attributed to Manifold Grace.
Network and firewall virtualisation
Security and privacy guidance
UK NCSC small organisation security guidance · NCSC online security basics
Virtual and local compute
Colab · Ollama · VirtualBox
Further official references for virtual machines, free tiers and accelerator limits
Microsoft Azure VM quickstart · Google Cloud Free Program and exclusions · Google Cloud GPU quotas and restrictions · GitHub Codespaces allowances.
Optional images that would improve coverage
Useful supplementary references are a labelled fibre ONT installation, a home switch/AP topology, a backup-and-restore comparison, a virtual firewall lab screenshot, a local-versus-hosted GPU cost model, and a clean network segmentation diagram. None is needed before using this page.
Status: technical reference and project exploration. No physical hardware, security certification, deployed AI workflow or completed research experiment is implied.
Browse 58 free and limited-free infrastructure, AI, cybersecurity and automation service links — categorized by role, free-status and example combinations.