I seem to remember that for production deployments...
# ask-ai
s
I seem to remember that for production deployments via ECS, there was something about using non-burstable or burstable something or other. please advise
t
Thanks for reaching out, Jason. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.
When deploying LiveKit on AWS ECS for production, it's really best to choose non-burstable instance types instead of burstable ones like T3 or T4g. Burstable instances use a CPU credit system, which can sometimes cause CPU throttling if the load is steady, leading to delays and timeout problems. This can be especially tricky for CPU-heavy tasks. To keep things running smoothly, I recommend going with compute-optimized non-burstable instances such as
m5
,
c5
,
c6i
, or
c7i
. These give you steady and reliable CPU performance, which is ideal for LiveKit’s real-time media and AI processing needs. If you do need to use burstable instances, it’s a good idea to keep an eye on the
CPUCreditBalance
metric in CloudWatch to make sure you don’t run out of credits. Also, enabling
t3.unlimited
mode can help, but just a heads-up—it might end up costing about the same as non-burstable instances. Sources: Troubleshooting Latency and Timeout Errors with Turn Detection on AWS | LiveKit turn detector plugin | LiveKit Docs | Docs::Home | Deploying LiveKit | LiveKit Docs