Numbers Every Programmer Should Know

This page is made for memorization.

Latency Numbers

Every programmer should know

A learning tool to help memorize napkin math. Reference: sirupsen/napkin-math

Network RTT tiers (universal): 20 / 40 / 60 / 120 / 200 ms — within country, nearby region, neighboring region, far region, very far. India and US example routes share the same tier cells.

Each grid is arranged by multiples of 1, 2, 5 across the columns and multipliers 1, 10, 100 down the rows. One grid per unit — ns, μs, ms, s — from fastest to slowest. Colors track the entry category, and overall hotter shades (reds) mark the fastest hardware, cooling toward blues and purples as latency climbs.

Warehouse queries (very high level).Snowflake query times depend on warehouse size, data volume, caching, and query plan. Entries in the ms and s grids are napkin anchors, not benchmarks.

ns

	1	2	5
1	Sequential Memory R/W (64 bytes)0.5 ns	Branch misprediction3 ns
10	Hashing, not crypto-safe (64 bytes)10 ns	Random Memory R/W (64 bytes)20 ns	Virtual thread create50 ns
100	HashMap get100 ns Virtual thread context switch100 ns Hashing, crypto-safe (64 bytes)100 ns	System Call300 ns

CPUMemoryCPU / KernelJava in-process

μs

	1	2	5
1	Sequential SSD read (8 KiB)1 μs JSON serialize 1 KB1 μs Log line (sync)1 μs	Sequential SSD write, -fsync (8 KiB)2 μs
10	Context Switch10 μs Platform thread context switch10 μs		TCP Echo Server (32 KiB)50 μs Proxy: Envoy/ProxySQL/Nginx/HAProxy50 μs
100	Random SSD Read (8 KiB)100 μs Spring MVC dispatch100 μs	Network within same region250 μs Premium network within zone/VPC250 μs Sequential SSD write, +fsync (8 KiB)300 μs	{MySQL, Memcached, Redis, ...} Query500 μs

CPU / KernelSSDDatabase / CacheNearby NetworkNetworkJava in-processJava frameworkJava JVM

ms

	1	2	5
1	JDBC pool checkout1 ms Platform thread create1 ms	JDBC round-trip (no pool)2 ms
10	Sequential HDD Read (8 KiB)10 ms Young GC pause10 ms Random HDD Read (8 KiB)10 ms	Network within India20 ms NA Central ↔ East25 ms Blob Storage GET, if-not-match 30430 ms	India ↔ Singapore40 ms NA Central ↔ West40 ms India ↔ Dubai60 ms NA East ↔ West60 ms
100	Blob Storage GET, 1 conn (128 KiB)80 ms Blob Storage GET, n conn (offsets)80 ms EU West ↔ NA East80 ms Blob Storage LIST100 ms India ↔ Europe120 ms	NA West ↔ Singapore180 ms Blob Storage PUT, 1 conn (128 KiB)200 ms Blob Storage PUT, n conn (multipart)200 ms India ↔ US240 ms Snowflake: simple query250 ms	Full GC pause500 ms Snowflake: moderate / less optimized500 ms

HDDObject StorageWithin countryNearby regionNeighboring regionFar regionVery farJava frameworkJava JVMWarehouse query

s

	1	5
1	Snowflake: heavy query (typical)1 s	Snowflake: heavy query (bad)5 s
10
100

Warehouse query

Latency Quiz

Test your knowledge

1 / 10Score: 0

What operation has this latency?

5 × 10⁻¹ s

Bandwidth

Throughput and transfer time (÷8)

1 Gbps ≈ 125 MB/s. 1 GB over 1 Gbps = 8 seconds. Same fact, two angles.

Link speed	Throughput (÷8)	1 GB transfer time	Mnemonic
1 Mbps	125 KB/s	8000 s	÷8
10 Mbps	1.25 MB/s	800 s	÷8
100 Mbps	12.5 MB/s	80 s	÷8
1 Gbps	125 MB/s	8 s	÷8
10 Gbps	1.25 GB/s	0.8 s	÷8

Bits to bytes: divide link speed by 8. 1 Gbps ≈ 125 MB/s. 1 GB over 1 Gbps = 8 seconds.

Bandwidth Quiz

Test your ÷8 reflex

1 / 5Score: 0

What link speed transfers 1 GB in this time?

0.8 s

Cloud Pricing

Approximate monthly costs

What	Amount	$ / Month
CPU	1	$15
GPU	1	$5000
Memory	1 GB	$2
Storage
├Warehouse Storage	1 GB	$0.02
├Blob (S3, GCS)	1 GB	$0.02
├Zonal HDD	1 GB	$0.05
├Ephemeral SSD	1 GB	$0.08
├Regional HDD	1 GB	$0.10
├Zonal SSD	1 GB	$0.20
├Regional SSD	1 GB	$0.35
Networking
├Same Zone	1 GB	$0
├Blob	1 GB	$0
├Ingress	1 GB	$0
├L4 LB	1 GB	$0.008
├Inter-Zone	1 GB	$0.01
├Inter-Region	1 GB	$0.02
├Internet Egress †	1 GB	$0.10
CDN Egress	1 GB	$0.05
CDN Fill ‡	1 GB	$0.01
Warehouse Query	1 GB	$0.005
Logs/Traces ♣	1 GB	$0.50
Metrics	1000	$20
EKM Keys	1	$1

Cost Quiz

Test your pricing knowledge

1 / 10Score: 0

What is the monthly cost?

Zonal SSD

Availability Table

Rule-of-thumb downtime to remember

Availability	Downtime per year (approx.)	Mnemonic	Meaning
90% (1 nine)	36.5 days	1 month	The service is down for a month.
99% (2 nines)	3.65 days	Long weekend	Down for a long weekend.
99.9% (3 nines)	8.76 hours ≈ 500 min	500	Down for a workday (~9 hours).
99.99% (4 nines)	52.6 minutes ≈ 50 min	50	Down for a lunch break (~50 minutes).
99.999% (5 nines)	5.26 minutes ≈ 5 min	5	Down for a coffee break (~5 minutes).
99.9999% (6 nines)	31.5 seconds ≈ 0.5 min	0.5	Down for half a minute.
99.99999% (7 nines)	3.15 seconds	3 sec	Down for a blink (~3 seconds).

Availability Mnemonics Quiz

Test the mnemonics

1 / 7Score: 0

Which availability level does this describe?

Down for a coffee break (~5 minutes).

Threads: Platform vs Virtual

Java concurrency napkin math

Metric	🧵 Platform Threads (OS-level)	🍃 Virtual Threads (Java 21+)
Memory overhead	~1 MB per thread (stack-allocated)	~2 KB to 10 KB per thread (heap-allocated)
Creation time	~1 to 2 ms (OS kernel system call)	~10 to 50 ns (Java object allocation)
Max capacity	~1,000 to 5,000 per server (OOM risk)	Millions on a standard laptop
Context switch cost	~1 to 10 μs (heavy OS operation)	~100 ns (managed by the JVM)

Platform threads map 1:1 to OS threads — heavy and scarce. Virtual threads (Java 21+) are cheap JVM-managed tasks, so you can run millions concurrently instead of a few thousand.

Power of 2

Data volume in distributed systems boils down to basics: a byte is a sequence of 8 bits; an ASCII character uses one byte of memory. Units are powers of 2: 2¹⁰ = 1024 ≈ 10³ = 1000 (1 Thousand), and so on.

Reference: ByteByteGo — Back of the Envelope Estimation

Power (2ⁿ)	Power (10ⁿ)	Approximate value	Full name	Short name
2¹⁰	10³	1 Thousand	Kilobyte	KB
2²⁰	10⁶	1 Million	Megabyte	MB
2³⁰	10⁹	1 Billion	Gigabyte	GB
2⁴⁰	10¹²	1 Trillion	Terabyte	TB
2⁵⁰	10¹⁵	1 Quadrillion	Petabyte	PB
2⁶⁰	10¹⁸	1 Quintillion	Exabyte	EB
2⁷⁰	10²¹	1 Sextillion	Zettabyte	ZB
2⁸⁰	10²⁴	1 Septillion	Yottabyte	YB

Power of 2 Quiz

Test the units and powers

1 / 8Score: 0

What is the decimal power?