System Design Interview Prep 2026: The Complete Roadmap — URL Shortener to Netflix-Scale Architecture (With Real Questions Asked at Google, Meta & Amazon)

After analyzing interviews at 40+ top tech companies, MockExperts has identified a universal framework that works across every system design problem — from simple URL shorteners to planet-scale social feeds.

1. Step 1 — Clarify Requirements (5–8 min): Define functional requirements (what the system must do) and non-functional requirements (scale, latency SLAs, availability targets, consistency model). Ask: "Are we designing for 10K users or 10 million?" and "Does this require strong consistency or eventual consistency?"
2. Step 2 — Capacity Estimation (3–5 min): Estimate daily active users (DAU), requests per second (RPS), data volume (GB/day), and bandwidth. Use these to justify technology choices downstream. Example: 100M DAU × 10 requests/day = ~11.5K RPS peak.
3. Step 3 — High-Level Design (8–10 min): Draw the macroscopic architecture: clients, CDN/load balancer, API gateway, application servers, databases, and caches. Keep this intentionally simple — no micro-optimizations yet.
4. Step 4 — API Design (3–5 min): Define the core REST/GraphQL endpoints or gRPC methods. This forces precision about what the system actually does and what data flows in/out.
5. Step 5 — Deep Dive on Critical Components (10–15 min): The interviewer will ask you to zoom into the hardest parts. Typical deep dives: database schema, sharding strategy, cache invalidation policy, rate limiting, or failure recovery.
6. Step 6 — Identify Bottlenecks & Scalability (5 min): Proactively identify single points of failure (SPOF), scaling bottlenecks, and operational edge cases. Propose mitigation strategies for each.

Walkthrough: Designing a URL Shortener (Bit.ly at Scale)

The URL shortener is the canonical entry-level system design question asked at nearly every major company. Mastering this question teaches the fundamentals: key generation, database selection, caching, and horizontal scaling.

Functional Requirements:

• Given a long URL, generate a unique 6–8 character short code.
• Redirect short URL to original URL with minimal latency (p99 < 50ms).
• Custom aliases (e.g., bit.ly/my-brand).
• Link expiry and click analytics.

Scale Estimation:

• 100M new URLs created per day → 1,160 writes/sec
• 10:1 read/write ratio → 11,600 redirects/sec
• Average URL: 500 bytes → 50GB storage/day → ~18TB/year

Key Generation Strategy:

The most critical design decision is the short code generation approach:

The Recommended Architecture:

Use a distributed counter with Snowflake-style ID generation (epoch timestamp + machine ID + sequence) encoded in Base62. Store mappings in Cassandra (optimized for high-write, key-value lookups at scale). Layer Redis with LRU eviction to cache the hottest 20% of links (which typically generate 80% of traffic). Use a CDN edge cache layer to serve redirects at <5ms globally.

Walkthrough: Designing Netflix — Video Streaming at Global Scale

This is the canonical senior-level question (L5/L6 at Netflix, Meta, YouTube). The key insight interviewers probe: video streaming is a fundamentally different problem from text/JSON API serving — it's a bulk data delivery problem requiring CDN-first design.

The 3 Core Subsystems:

1. Ingest & Transcoding Pipeline: Videos uploaded by creators pass through a distributed transcoding farm (AWS Elemental, or custom NVENC GPU clusters) that generates 15–20 resolution profiles (4K/1080p/720p/480p/360p) × multiple audio codecs (Dolby Atmos, AAC, Opus) × multiple bitrate variants per profile. Output segments are stored in S3-compatible object storage.
2. Adaptive Bitrate (ABR) Streaming: The player does not download a static file. It downloads 2–4 second MPEG-DASH or HLS segments and dynamically switches quality tiers based on real-time bandwidth probing. The manifest file (.m3u8 or .mpd) tells the player which segment URL to fetch next.
3. CDN Architecture: Netflix operates its own CDN called Open Connect, placing specialized OCA (Open Connect Appliance) servers inside ISP data centers. The top 10,000 popular titles are pre-positioned on edge nodes nightly. User requests are geo-routed to the nearest OCA. Less popular content falls back to Amazon CloudFront or S3 direct.

Top 10 System Design Questions Asked in 2026

How to Score in the Top 10% on System Design Interviews

Based on our analysis of 1,200+ recorded system design mock interviews at MockExperts, here are the behaviors that separate top 10% scorers from the median:

1. They state trade-offs explicitly: Rather than saying "I'll use Cassandra," top candidates say "I'll use Cassandra because its LSM-tree storage engine optimizes write throughput at the cost of slightly slower reads — which matches our 10:1 write-heavy workload."
2. They proactively identify failure modes: Without being asked, they enumerate: "The key generation service is a SPOF. I'd run it in active-active across 3 AZs with Zookeeper-based leader election for the counter segment assignments."
3. They scope smartly: They tell the interviewer what they're deprioritizing and why — "I'm going to skip the analytics pipeline for now and focus on the hot path. Should I come back to it in 5 minutes?"
4. They draw before they speak: The best system design candidates start sketching components within 60 seconds of understanding the problem. Visuals anchor the conversation and prevent getting lost in abstract discussions.