#Infrastructure
380 posts
Aurora vs RDS: The Structural Difference and When to Switch
A working standard for choosing between Aurora and standard RDS: what the separated compute/storage architecture buys (15 shared-storage read replicas, fast failover, 256TiB auto-scaling), the three-layer pricing where storage and I/O ride on an instance premium, the 25% threshold that makes I/O-Optimized cheaper, Serverless v2 with scale-to-zero, and the conditions where standard RDS remains the right answer.
When HPA Doesn't Behave — 6 Common Configuration Mistakes
Why a Kubernetes HPA fails to scale, scales too late, or oscillates all day: missing requests and what the target percentage really means, scale-up reaction time and stabilization windows, flapping, GitOps fighting over replicas, and the limits of scaling on CPU alone.
S3 Storage Classes Compared: Standard to Glacier Deep Archive
A working standard for choosing S3 storage classes: the full map of storage rates versus retrieval fees and minimum durations, the three traps in Standard-IA (retrieval fees, 30 days, 128KB), the real differences between the three Glacier classes (instant, hours, 12+ hours), why Intelligent-Tiering is the answer when access patterns are unknown, and when lifecycle transition fees eat the savings.
When a Pod Is Stuck in Pending — 7 Causes and the Order to Check Them
The causes of Kubernetes Pods stuck in Pending, ordered by real-world frequency: reading the Events in describe, insufficient resources, oversized requests, nodeSelector and affinity mismatches, taints, PVC binding waits, ResourceQuota, and the cases where Cluster Autoscaler refuses to scale up.
ECS Fargate vs EC2: Spec Combinations and Price Compared
A working standard for choosing ECS compute: the structural difference between per-task billing (Fargate) and per-instance billing (EC2), the vCPU/memory grid Fargate allows (0.25-16 vCPU, up to 120GB), what the premium actually amounts to at equal specs, the point where utilization and bin packing flip the comparison, and the levers that cut the bill — ARM, Spot, and Savings Plans.
Sizing Servers for Self-Hosted LLMs — From VRAM Math to Concurrent Users
A repeatable order for estimating the GPU scale a self-hosted LLM needs: computing VRAM from parameter count and precision, the KV cache and its relationship to concurrency, memory bandwidth and token speed, example configurations by scenario, and the break-even against API pricing.
AI Training Servers vs. Inference Servers — Where the Requirements Diverge
Why the same "AI server" question has two very different answers: batch versus request-serving workload shapes, GPU interconnect versus memory bandwidth as the bottleneck, precision and quantization, cost structures, and the trap of trying to share one fleet for both.
RDS Instance Classes Compared: db.t vs db.m vs db.r
A working standard for choosing RDS instance classes: the db. prefix on EC2 naming rules, why the same hardware costs roughly twice as much (management included, Multi-AZ extra), why db.t does not belong in production, the buffer-pool criterion that separates db.m from db.r, and why moving to Graviton is easier on RDS than on EC2.
EC2 Instance Families Compared: Choosing Between t, m, c, and r
A working standard for choosing EC2 instance types: how to read type names (family, generation, attributes), when the t burstable model fits and when it backfires, the vCPU-to-memory ratios behind m, c, and r with large-size price comparisons, choosing between Graviton, Intel, and AMD, and sizing from CloudWatch and Compute Optimizer measurements.
How to Read NVIDIA GPU Names — Architecture Generations and Product Lines
Reading H100, B200, GB300, L40S, and RTX 5090 as a system rather than a code: the scientist-named architecture axis and the datacenter/inference/consumer product-line axis, what the digits and prefixes mean, and the generation landscape and roadmap as of mid-2026.
GPU Servers vs. General Servers — Architecture, Power, Cooling, Price
Where a GPU server stops being a regular server with a graphics card in it: the inverted architecture and interconnects, power density an order of magnitude higher and the arrival of liquid cooling, the price structure, the driver stack, and why utilization management becomes the job.
gp3 vs io2 — How to Choose an EBS Volume Type
A working standard for choosing EBS volume types: the legacy of gp2's size-coupled, burst-credit model, gp3's independent provisioning and raised ceilings (80,000 IOPS, 2,000MB/s), the conditions that actually justify io2 Block Express, sizing IOPS from measurements, and converting gp2 to gp3 with no downtime.