Sizing and capacity - Hilt Documentation

Measured resource usage

Host type	RSS	CPU	Filtered noise	Events/s shipped
`m6i.large` general-purpose VM	4–6 MB	< 1% of 1 core	~60–75%	50–400
`r6i.xlarge` DB host	6–8 MB	~1% of 1 core	~70–80%	300–1,500
EKS worker (16 pods, busy)	5–7 MB	< 2% of 1 core	~70%	800–3,000

Defaults

The collector ships with conservative defaults suitable for most environments:

Kubernetes: the DaemonSet requests 100m CPU / 128Mi memory with limits of 500m / 512Mi.
VMs: no hard limits are applied by default, though they can be enforced during installation (e.g., CPUQuota=20% and MemoryMax=512M) if a strict ceiling is required.

Ring buffer tuning

The default ring buffer size is 1 MB per CPU, which is sufficient for workloads up to ~5,000 events/sec per CPU. For high-throughput systems, increasing this to 8–16 MB per CPU is recommended. The agent degrades gracefully under load: when userspace cannot drain the ring buffer fast enough, the eBPF side drops oldest events and emits a count of dropped events to the metrics stream (metrics.c). Dropped events never cause the agent to block the kernel.