Calculate packets-per-second rate and its CPU load impact on network interrupt handling.
Line-rate packets-per-second isn't simply bandwidth divided by frame payload size — the wire-level cost of every Ethernet frame also includes a fixed preamble/SFD (used for clock synchronization) and a mandatory interframe gap (idle time between frames), both of which consume link time without carrying data. Dividing total link bandwidth by the sum of frame size, preamble, and IFG (all converted to bits) gives the maximum achievable packet rate — this is why small packets achieve far lower throughput efficiency than large ones: the fixed overhead is amortized over less payload.
Packets per second
pps = bandwidth_bps / (frame_size_bits + IFG_bits + preamble_bits)
Vendors quote 'line-rate' or 'wire-speed' PPS using the smallest legal Ethernet frame (64-byte payload, 84 bytes on the wire with overhead) because that's the worst case for a switch/router's per-packet processing budget — a device rated for line-rate at 64 bytes can handle any larger, less demanding frame size at full bandwidth too.
`sar -n DEV 1` reports rxpck/s and txpck/s per interface per second; `ethtool -S <iface>` on supported NICs exposes hardware packet counters that can be sampled and diffed for precise rate calculation.
Yes — larger frames (e.g. 9000-byte jumbo frames) amortize the fixed preamble/IFG overhead over more payload bytes, reducing the packets-per-second needed to sustain the same throughput and lowering per-packet CPU/interrupt overhead, at the cost of requiring jumbo-frame support across the entire path.