Slow server simulator
This repo contains a very simple simulator for exploring the behavior of a slow server in a simple distributed system. For more information, see the accompanying blog post.
The configuration of the system is currently hardcoded. It consists of:
- a time limit (how long a period to simulate)
- a total level of client concurrency
- an array of server classes, each of which includes a count and latency-per-request for that server.
The hardcoded configuration runs a 10-minute simulation with 10,000 clients, 1
"slow" server that takes 1s per request, and 99 "fast" servers that take 1ms per
request. Results (server_0 is the "slow" server):
$ node sim.js
simulated time: 600000 ms
total client concurrency: 10000
servers:
1 server that completes requests with latency 1000 ms (starting with "server_0")
99 servers that complete requests with latency 1 ms (starting with "server_1")
server_0 5455039 requests ( 9092 rps)
server_1 5462256 requests ( 9104 rps)
server_2 5459316 requests ( 9099 rps)
server_3 5463211 requests ( 9105 rps)
...
server_96 5459207 requests ( 9099 rps)
server_97 5458421 requests ( 9097 rps)
server_98 5458234 requests ( 9097 rps)
server_99 5456471 requests ( 9094 rps)
overall: 545829375 requests (909715 rps, expect 9097 rps per server)
As a sanity check on the simulator, if we change the simulation time to 990ms,
server_0 completes no requests (as expected), while other servers complete
many:
$ node sim.js
simulated time: 990 ms
total client concurrency: 10000
servers:
1 server that completes requests with latency 1000 ms (starting with "server_0")
99 servers that complete requests with latency 1 ms (starting with "server_1")
server_0 0 requests ( 0 rps)
server_1 9954 requests ( 10055 rps)
server_2 10114 requests ( 10216 rps)
...
server_97 9844 requests ( 9943 rps)
server_98 10072 requests ( 10174 rps)
server_99 10031 requests ( 10132 rps)
overall: 995406 requests (1005460 rps, expect 10054 rps per server)
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent