Benchmarks & Profiling

Disaggregated memory pool bandwidth, RPC latency, and failover timing

Measurements in Preparation

Formal real-device benchmark measurements for AMEVA-Cluster require a minimum two-device fleet configuration (one master, at least one worker). Measurements are being prepared on the Tailscale-connected Galaxy S25 (master) + Galaxy A35 / A53 (workers) fleet.

Planned measurement axes: TCP memory-read throughput (GB/s), RPC round-trip latency (ms), HMAC-SHA256 Guard authentication overhead (µs), UFS 3.1 flash failover switch time (ms), and peak worker-node RSS overhead (MB).

Planned Measurement Conditions

MetricPlanned Fleet ConfigConditionsStatus
TCP memory-read throughputS25 (master) + A35 (worker)Tailscale LAN, 4KB–4MB block sizes, 1 GB transferPending
RPC round-trip latencyS25 (master) + A35 (worker)Loopback vs Tailscale vs Wi-Fi direct, 100 iterationsPending
HMAC Guard authentication overheadS25 standaloneHMAC-SHA256, 1000 verificationsPending
UFS 3.1 flash failover switch timeS25 (master)Simulated worker disconnect, mmap streaming activationPending
Worker-node RSS overheadA35 / A53 (worker-only)Idle RAM-cache mode, 300MB guard-band activePending