Cloud Storage & Virtualization Explained | Draftsbook

Lesson Focus & Objectives

Virtualization is the fundamental technology that makes cloud computing possible. By decoupling computing resources from physical silicon, cloud providers achieve astronomical server utilization rates through dynamic multi-tenant slicing. In Part 6, we analyze Type-1 bare-metal hypervisors versus Type-2 hosted hypervisors, explore modern microVMs and container runtimes, and dissect the three fundamental cloud storage architectures: Block Storage, Object Storage, and Network File Storage.

1. Hypervisors: Type-1 vs. Type-2 vs. MicroVMs

Type-1 Bare-Metal

Runs directly on the host's physical silicon without an intervening host operating system (e.g., VMware ESXi, Linux KVM, Xen). Provides direct hardware execution, minimal hypervisor CPU overhead (< 2%), maximum I/O performance, and complete multi-tenant hardware isolation. Powers all hyperscale public cloud providers.

Type-2 Hosted

Runs as an application inside a conventional host operating system (e.g., Oracle VirtualBox, VMware Workstation). Every I/O interrupt and memory page fault traverses both the hypervisor and the underlying host kernel, incurring significant latency. Primarily used for local workstation testing.

MicroVMs & Containers

Technologies like AWS Firecracker and Kata Containers combine the near-zero overhead and millisecond boot speed of Linux cgroups/namespaces with the cryptographic hardware boundary of hardware virtualization, powering serverless computing (AWS Lambda, Fargate).

2. The Three Cloud Storage Archetypes

Storage Tier	Access Protocol & Abstraction	Latency / IOPS	Primary Use Case
Block Storage	Raw 512-byte / 4KB sectors via iSCSI or NVMe-oF (e.g., AWS EBS, Azure Disk)	Sub-millisecond latency; up to 256,000 IOPS	High-performance relational databases (PostgreSQL, MySQL), VM OS boot volumes
Object Storage	Flat namespace, HTTP REST API (GET/PUT), metadata tags (e.g., AWS S3, GCS)	High latency (50-100ms TTFB); unbounded petabyte throughput	Static assets, media streaming, backups, machine learning data lakehouses
File Storage	Hierarchical POSIX filesystem via NFS v4 or SMB (e.g., AWS EFS, Azure Files)	Low latency (1-5ms); shared multi-client concurrent mounts	Legacy CMS uploads (WordPress), shared development workspaces, multi-node compute jobs

3. Cloud Resource Provisioning & Auto-Scaling

Cloud management engines dynamically maintain Service Level Agreements (SLAs) through automated feedback control loops:

Horizontal Pod Autoscaler (HPA): Samples CPU, memory, or custom queue metrics (e.g., RabbitMQ queue length) every 15 seconds to scale replica counts dynamically.
Cluster Autoscaler: Detects pending pods that cannot be scheduled due to resource constraints and calls cloud APIs to provision new physical worker nodes.
Over-Provisioning & Ballooning: Hypervisors dynamically allocate memory pages from idle virtual machines to active workloads using memory balloon drivers, maximizing hardware ROI.