Database indexing in HLD - Scalability & System Analysis

Scalability Analysis - Database indexing

Growth Table: Database Indexing at Different Scales

Users / Queries	100 Users	10,000 Users	1 Million Users	100 Million Users
Query Volume (QPS)	~100 QPS	~1,000 QPS	~10,000 QPS	~100,000 QPS
Data Size	MBs to low GBs	GBs to tens of GBs	Hundreds of GBs to TBs	Multiple TBs to PBs
Index Size	Small, fits in memory	Larger, partial in memory	Large, needs optimized storage	Very large, requires partitioning
Query Latency	Low (ms)	Low to moderate	Moderate to high if unoptimized	High without advanced indexing
Index Maintenance Cost	Low	Moderate	High, impacts write speed	Very high, requires tuning

First Bottleneck

As user count and query volume grow, the first bottleneck is usually the database query performance. Without proper indexing, queries become slow because the database scans large tables. Indexes speed up reads but add overhead on writes. At medium scale, index maintenance and memory usage become bottlenecks. At very large scale, index size and disk I/O limit performance.

Scaling Solutions for Database Indexing

• Index Optimization: Use appropriate index types (B-tree, hash, composite) based on query patterns.
• Partial and Covering Indexes: Index only needed columns to reduce size and improve speed.
• Index Maintenance: Schedule index rebuilds and updates during low traffic to reduce write impact.
• Horizontal Scaling: Shard data to distribute indexes across multiple database instances.
• Caching: Use in-memory caches (Redis, Memcached) to reduce database hits for frequent queries.
• Read Replicas: Offload read queries to replicas with their own indexes to reduce load on primary.
• Partitioning: Split large tables and indexes by range or hash to improve query and maintenance performance.

Back-of-Envelope Cost Analysis

Assuming 10,000 QPS at medium scale:

• Each query reads indexed data in ~1-5 ms.
• Index size can be 10-30% of data size; for 1 TB data, index ~100-300 GB.
• Memory needed to hold hot indexes: tens of GBs to hundreds of GBs.
• Write overhead: each insert/update requires index update, increasing write latency by ~10-30%.
• Network bandwidth depends on replication and sharding; expect 1-10 Gbps for large scale.

Interview Tip

When discussing database indexing scalability, start by explaining how indexes improve read speed but add write overhead. Then describe how index size and maintenance become bottlenecks as data grows. Finally, outline solutions like index optimization, partitioning, caching, and sharding. Use clear examples and relate to real-world query patterns.

Self Check Question

Your database handles 1000 QPS with indexes. Traffic grows 10x to 10,000 QPS. What do you do first and why?

Answer: The first step is to analyze if the current indexes and hardware can handle the increased read load. Likely, the database CPU or disk I/O will become bottlenecks. To fix this, add read replicas to distribute read queries and reduce load on the primary. Also, consider caching frequent queries to reduce database hits. This approach improves read scalability without immediately changing indexes or hardware.

Key Result

Database indexing improves read speed but adds write overhead; as data and queries grow, index size and maintenance become bottlenecks requiring optimization, partitioning, caching, and horizontal scaling.

Practice

1. What is the primary purpose of a database index?

easy

A. To increase the size of the database

B. To speed up data retrieval by allowing quick lookups

C. To store backup copies of data

D. To encrypt the database contents

Solution

1. Step 1: Understand what an index does

   An index helps the database find data faster by creating a quick lookup structure.

2. Step 2: Identify the main benefit

   This quick lookup reduces the time needed to search large tables.

3. Final Answer:

   To speed up data retrieval by allowing quick lookups -> Option B

4. Quick Check:

   Index = Faster data search [OK]

Hint: Indexes speed up searches by creating shortcuts [OK]

Common Mistakes:

• Thinking indexes store extra data copies
• Confusing indexes with backups
• Assuming indexes encrypt data

2. Which of the following is the correct syntax to create an index on the column user_id in SQL?

easy

A. CREATE INDEX idx_user ON users (user_id);

B. INDEX CREATE user_id ON users;

C. CREATE users INDEX ON user_id;

D. MAKE INDEX user_id IN users;

Solution

1. Step 1: Recall SQL syntax for creating indexes

   The standard syntax is CREATE INDEX index_name ON table_name (column_name);

2. Step 2: Match the syntax with options

   CREATE INDEX idx_user ON users (user_id); matches the correct syntax exactly.

3. Final Answer:

   CREATE INDEX idx_user ON users (user_id); -> Option A

4. Quick Check:

   CREATE INDEX ... ON ... (column) [OK]

Hint: Remember: CREATE INDEX name ON table(column) [OK]

Common Mistakes:

• Swapping table and column names
• Using wrong keyword order
• Missing parentheses around column

3. Given a table orders with an index on order_date, what will be the output of the query plan for SELECT * FROM orders WHERE order_date = '2023-01-01';?

medium

A. Full table scan without using the index

B. Ignoring the order_date column completely

C. Sorting all rows by order_date before filtering

D. Index seek on order_date to quickly find matching rows

Solution

1. Step 1: Understand how indexes affect query plans

   An index on order_date allows the database to quickly locate rows matching the date.

2. Step 2: Identify the expected query plan

   The database will perform an index seek to find matching rows efficiently, not scan the whole table.

3. Final Answer:

   Index seek on order_date to quickly find matching rows -> Option D

4. Quick Check:

   Indexed column = index seek [OK]

Hint: Indexed WHERE columns use index seek, not full scan [OK]

Common Mistakes:

• Assuming full table scan despite index
• Confusing sorting with filtering
• Ignoring index usage in query plan

4. A developer created an index on a column but notices no improvement in query speed. What could be a likely reason?

medium

A. The database automatically disables all indexes

B. The index was created with the wrong SQL syntax

C. The queries do not filter or sort using the indexed column

D. Indexes always slow down queries

Solution

1. Step 1: Analyze index usage conditions

   Indexes only help if queries use the indexed column in WHERE, JOIN, or ORDER BY clauses.

2. Step 2: Identify why no speed improvement occurs

   If queries don't use the indexed column, the index is ignored, so no speed gain.

3. Final Answer:

   The queries do not filter or sort using the indexed column -> Option C

4. Quick Check:

   Index helps only if column used in query [OK]

Hint: Index unused if queries ignore indexed columns [OK]

Common Mistakes:

• Assuming syntax error disables index
• Believing all indexes slow queries
• Thinking database disables indexes by default

5. You have a large table with frequent inserts and queries filtering by status and created_at. Which indexing strategy balances read speed and write overhead best?

hard

A. Create a composite index on (status, created_at)

B. Create separate indexes on status and created_at individually

C. Create no indexes to avoid write overhead

D. Create a full-text index on status and created_at

Solution

1. Step 1: Understand composite vs separate indexes

   A composite index on (status, created_at) helps queries filtering on both columns efficiently.

2. Step 2: Consider write overhead and query patterns

   Separate indexes increase write overhead more than one composite index and may be less efficient for combined filters.

3. Step 3: Exclude unsuitable options

   No indexes hurt read speed; full-text index is for text search, not filtering by status/date.

4. Final Answer:

   Create a composite index on (status, created_at) -> Option A

5. Quick Check:

   Composite index = balanced read/write [OK]

Hint: Composite indexes optimize multi-column filters with less overhead [OK]

Common Mistakes:

• Creating many separate indexes causing slow writes
• Skipping indexes hurting read performance
• Using full-text index for non-text filtering