Redis is a very popular in-memory store and has support for Lua scripts. But how about throwing LuaJIT into the mix?
This is a quick post about my most recent endeavor, adding LuaJIT support to redis.
Porting the existing Lua EVAL command to use LuaJIT was surprisingly straightforward. You can check out the code here.
It’s not production-ready, but the performance gains for compute-heavy lua code are obvious:
# luajit
$ redis-benchmark -n 1000 EVAL "local function fib(n)if n<2 then return n end return fib(n-1)+fib(n-2)end return fib(20)" 0
Summary:
throughput summary: 4830.92 requests per second
latency summary (msec):
avg min p50 p95 p99 max
10.035 1.280 10.175 10.903 11.183 11.399
# lua
redis-benchmark -n 1000 EVAL "local function fib(n)if n<2 then return n end return fib(n-1)+fib(n-2)end return fib(20)"
Summary:
throughput summary: 1904.76 requests per second
latency summary (msec):
avg min p50 p95 p99 max
25.841 3.184 25.711 27.631 28.047 28.511
TL;DR: 2.5x increase in throughtput and 2.5x decrease in latency
While this doesn’t necessarily translate to real-world production workloads, it clearly highlights the performance improvements LuaJIT can offer over standard Lua.