System load is the amount and kind of work a system must handle: request rates, data volume, concurrent users, and operations such as reads or writes. It differs from throughput, which measures requests or data processed per second, and scalability, which describes handling increases in load. section 2.2.4
A useful load profile accounts for changing demand and different workloads. Peaks may follow daily or seasonal patterns, campaigns, or external events; read-heavy paths, write-heavy paths, user interactions, and background processing can have different needs. Average load alone can hide the conditions that strain a system. section 2.2.4
Compare peak scenarios
Hundreds of thousands of requests per second with small payloads pose a different challenge from a few requests per second with gigabyte-sized payloads. Asking whether traffic can spike tenfold for an hour, or whether end-of-month processing can run while users are active, makes constraints clearer than a single requests-per-second figure. section 2.2.4
Table 2.4 Load tiers
|
Tier
|
Peak concurrency
|
Throughput
|
Architecture implications
|
|
Consumer storefront
|
50,000–100,000 concurrent users
|
5,000–15,000 req/s
|
Autoscaling, CDN, read replicas.
|
|
B2B
|
1500–2,000 concurrent users
|
100–500 req/s
|
Fixed capacity, vertical scaling sufficient
|
|
Internal backoffice
|
50–200 concurrent users
|
10–50 req/s
|
Single cluster, no autoscaling needed
|
|
Background processing
|
Low concurrency
|
High data volume, latency-tolerant
|
Batch jobs or stream processing, queue-based workers
|
Compare illustrative peak-load tiers: a consumer storefront, B2B service, internal backoffice, and background processing, each with distinct workload and capacity implications. section 2.2.4
Compare load alongside availability, durability, latency, and volatility before discussing solutions. A storefront search and a backoffice search may share a feature label yet have different service-tier needs and cost implications. section 2.2.6
