System Snapshot Comparison: Full Analysis#

System Overview#

All snapshots were recorded from a Debain Bookworm-based Raspberry Pi 5 (Cortex-A76, 4 CPU cores, 8GB RAM, NVMe SSD root), running a similar stack of media, storage, and proxy services. Below is a direct comparative analysis of all available logs, including before and during usage of Cloudflare Tunnel, tailscale, and Google Cloud reverse proxy routing.

Key Metrics Across All Snapshots#

MetricPrevious ReportsCloudflare Tunnel logGCP Reverse Proxy log
OS/PlatformDebian Bookworm, kernel 6.12.34, aarch64SameSame
CPU4x Cortex-A76SameSame
Memory (Total)8063 MB8063 MB8063 MB
Memory (Used)~2800–3040 MB~3000–3100 MB~3000 MB
Memory (Free)~320–400 MB250–310 MB280–297 MB
Swap Used2 MB2 MB2 MB
Disk / Root235 GB total, 153 GB used, 69% usageNo changeNo change
CPU Usage1–8% base, rare peaksMore bursty (up to 10–12%)More bursty (up to ~8%), some intervals 2–5%
Load Average0.13–0.200.18–0.31 (with peaks ~0.53)0.23–0.53 (with transient peaks)
Disk I/OVery light (rare spikes)Slightly elevated, more burstsMore bursts, but unsaturated
Top ProcessesMedia stack (Jellyfin, Sonarr, Radarr, etc), DockerSame plus Cloudflared roles more visibleCaddy, Cloudflared, Docker, etc.
NetworkNot detailed/normalTunnel: no bottlenecksGCP: not detail, but stable

Detailed Observations#

1. CPU & Memory Usage Patterns#

  • Base usage consistently low: Core system remains 94–98% idle in all logs.

  • Cloudflare/GCP Tunnels introduce occasional bursts: Tunnel daemons (cloudflared/tailscaled) and media processes (Jellyfin, etc.) show short CPU and RAM spikes, which do not lead to prolonged high utilization.

  • High process count: All logs show ~360–370 processes running, but only 1–2 in running state.

2. Disk and I/O#

  • Storage utilization unchanged: Disk usage stays at 69% of root NVMe, indicating no significant data growth.

  • I/O slightly more variable in tunnel logs: More frequent write spikes (up to 73–249 KB/s), but overall %util never approaches disk bottleneck (always <1% most of the time).

3. Memory and Swap#

  • Free memory slightly decreasing trend: From ~400 MB free in early reports, now hovers between 276–297 MB free with similar cache/buffer sizes. This is still healthy for an 8GB RAM system.

  • Swap essentially unused: All snapshots report 2 MB swap usage, indicating no memory pressure or thrashing.

4. Process Landscape#

  • Consistent stack: Docker, Jellyfin, Sonarr, Radarr, Prowlarr, Minio, Immich, Caddy, Nginx, Cloudflared, Tailscaled, Backup, etc. always visible.

  • Cloudflared and Tailscaled: Prominent only in tunnel snapshots, showing slightly increased %CPU during bursts.

  • No abnormal OOM-kills or zombie processes across any sample.

5. Load Average#

  • Generally low: Even during highest sampling, load averages top out at 0.53, well below even 1.0—meaning the 4-core system is rarely loaded more than ~1/8th of its full capability.

6. Network#

  • No bottlenecks in any sample: Outbound and inbound, the network is not under strain, with all services reporting healthy states.

Summary Table of Key Differences#

AreaEarly SnapshotsTunnel SnapshotsGCP Reverse Proxy SnapshotChange/Trend
CPUIdle, <8% typicalOccasional peaks up to 12%Occasional peaks to ~8%More frequent, short-lived bursts
RAM~2.8–3.0 GB used~3.0–3.1 GB used~3.0 GB usedSlightly higher, by ~200 MB
Disk I/OOccasional low spikesMore frequent, mid spikesFrequent small burstsMild increase, but not impactful
ProcessesMedia, Docker+Cloudflared prominent+Cloudflared, +CaddyOnly process profile varies
Swap2 MB2 MB2 MBNo change
Load Avg0.13–0.200.18–0.53 (peak at 0.53)0.23–0.53Small increases during load
Service ErrorsNoneNoneNoneAll remain healthy

Practical Implications#

  • No major resource or system health issues are present in any report.

  • Recent logs (Tunnel/Proxy) show modest, expected increases in CPU/memory/disk I/O volatility—a direct consequence of routing and encryption overhead, but still with ample resource headroom.

  • Service stack remains stable and responsive: No indications of contention or memory/CPU starvation, no services killed.

  • Performance headroom remains high: All critical metrics are well within safe operational ranges.

Conclusion#

The core difference when using Cloudflare or GCP tunnels is a slightly brisker pace of short-lived resource spikes, mainly in CPU and disk I/O, with a gradual minor reduction in free memory. Overall load, memory, and disk utilization remain healthy, with the server continuing to exhibit robust stability and reserve capacity for additional demand.

No critical bottlenecks are detected; small increases in system turbulence are expected and well within Raspberry Pi 5’s capabilities.