Hardware, workload, and scoring

Fixed application budget

Resource	Single-server layout	Four-server layout	Source
Application servers	1 × 8 vCPU, 16 GB RAM	4 × 2 vCPU, 4 GB RAM	3:44
Total application resources	8 vCPU, 16 GB RAM	8 vCPU, 16 GB RAM	3:44
Monthly application-server budget	$112	$112	3:44

• These are cloud vCPUs, not necessarily dedicated physical cores. Host differences can affect results. 4:06 5:16
• NGINX runs on a separate eight-core machine; Postgres runs on a managed 16-vCPU cluster. Both stay fixed and fall outside the $112 application budget. 4:06 4:26

Workload and pass criteria

• Virtual users browse articles, open profiles, load comments, and occasionally write. Three load-generator machines supply traffic; 1,000 users generate roughly 250–270 requests/second in this workload. 4:26
• The score is the maximum user count sustained for a full five-minute measurement window while meeting all three limits. 4:26 4:47
  1. 95% of requests take less than 500 ms. 4:26
  2. 99% take less than one second. 4:26
  3. Fewer than 1% fail. 4:26
• Shorter tests locate the approximate limit; a longer run confirms it. A brief spike does not qualify as sustained capacity. 4:47
• Read the results as workload-specific capacity under these latency and error limits. CPU time is a separate measure of work, not end-to-end response latency. 4:26 6:20 6:42