My team built a heavy duty load testing simulator in preparation for a upcoming launch. We kept dialing the simulated clients up until we finally saw timeouts talking to Redis (we used RedisLabs). The cause? We had saturated the 1 gig Ethernet link our AWS C4 instances had. Upgrading to the next tier of NICs solved the problem and Redis kept up the pace.
That said, Redis is hugely expensive for any meaningfully sized data set. Thankfully our use case allowed us to keep our data size bounded over time, or else I would’ve traded latency for drastically reduced cost (eg use Postgres).
That said, Redis is hugely expensive for any meaningfully sized data set. Thankfully our use case allowed us to keep our data size bounded over time, or else I would’ve traded latency for drastically reduced cost (eg use Postgres).