Architecture lab

Letter summaries on a bank’s own GPUsTarget the peak
0% busy
Mail intake
1x inst0 RPS
0% busy
Load balancer
1x inst0 RPS
0% busy
Summary API
2x inst0 RPS
0% busy
Model server
4x inst0 RPS

Design summary

Components

  • Mail intake: 1 × Users
  • Load balancer: 1 × Load balancer, 0 requests a second, 0% busy
  • Summary API: 2 × App server, 0 requests a second, 0% busy
  • Model server: 4 × Model server, 0 requests a second, 0% busy

Connections

  • Mail intake sends requests to Load balancer
  • Load balancer sends requests to Summary API
  • Summary API sends requests to Model server

0 of 0 requests a second served, average 5,997 ms, p99 12,578 ms, 5,515 USD a month.