Reading a datasheet like an engineer: throughput vs switching capacity
STANDARDS 08 APR 2026 7 MIN
Two numbers sit at the top of almost every switch datasheet: a switching capacity in gigabits per second, and a forwarding rate in millions of packets per second. They look like marketing flourishes. They are not. They describe two genuinely different limits, and a switch can be honest about one while quietly failing the other. Learning to read both — and the relationship between them — is the difference between trusting a spec sheet and being sold by it.
Switching capacity is a bandwidth ceiling
Switching capacity, sometimes called switching fabric or backplane bandwidth, is the total bits-per-second the switch can move when every port is busy. The convention is to count it as full-duplex and sum across all ports.
Take a 24-port gigabit switch with four 10GbE uplinks. The math the datasheet wants you to do:
- 24 ports × 1 Gbps × 2 (full duplex) = 48 Gbps
- 4 ports × 10 Gbps × 2 = 80 Gbps
- Total = 128 Gbps switching capacity
If the datasheet lists exactly that figure, the switch is non-blocking on bandwidth: the fabric can carry every port at line rate simultaneously. If the listed capacity is lower than the sum of the ports, the switch is oversubscribed by design — fine for many access roles, a problem in an aggregation role, and something you want to know before you buy, not after.
So the first check is simple arithmetic: add up the ports yourself and compare to the printed switching capacity. If they match or the printed number is higher, bandwidth is not your constraint. If the printed number is lower, find out by how much.
Forwarding rate is a packets-per-second ceiling
Bandwidth is not the only thing a switch runs out of. Every frame, regardless of size, costs the switching ASIC a fixed amount of work: a lookup, a decision, a forward. That work is counted in packets per second, and it has its own ceiling independent of how many bits those packets carry.
This is why the worst case is the smallest frame. The standard reference is the 64-byte minimum Ethernet frame. Add the mandatory overhead — preamble, start-of-frame delimiter, and interframe gap, 20 bytes total — and a minimum frame occupies 84 bytes, or 672 bits, on the wire.
For a single gigabit port at line rate:
- 1,000,000,000 bits/sec ÷ 672 bits/frame = 1,488,095 frames/sec
Round it: a gigabit link maxes at about 1.488 Mpps. A 10GbE link is ten times that, roughly 14.88 Mpps. To forward all 64-byte frames at line rate, our 24×1G + 4×10G example needs:
- (24 × 1.488) + (4 × 14.88) = 35.71 + 59.52 = 95.2 Mpps
If the datasheet’s forwarding rate is at or above 95.2 Mpps, the switch is non-blocking on packets too — it can handle a full load of minimum-size frames. If the forwarding rate is lower than that, the switch will drop packets under small-frame load even though its switching capacity says it has bandwidth to spare. That is the gap.
The gap is where datasheets hide
A switch can advertise a fully non-blocking 128 Gbps fabric and a forwarding rate well under 95 Mpps. Both numbers can be true. What it means is that the box can move big frames at full bandwidth but chokes on a flood of small ones — exactly the load you see with heavy VoIP, certain video, gaming, telemetry, or a small-packet denial-of-service.
So read the two numbers together:
- Compute the non-blocking bandwidth from the port count and compare to switching capacity.
- Compute the non-blocking forwarding rate at 64-byte frames using 1.488 Mpps per gigabit, and compare to the listed forwarding rate.
- If either listed number falls short of your computed figure, the switch is oversubscribed on that axis. Decide whether your traffic profile cares.
For a leaf or access switch carrying mostly large flows, a forwarding-rate shortfall may never bite. For an aggregation switch, a firewall-adjacent role, or anything exposed to small-packet load, it absolutely will.
A few honesty checks
- Watch the duplex convention. A vendor that quotes switching capacity in full-duplex but forwarding rate in a way that assumes large frames is comparing apples to oranges to make both look generous. Normalize both to the same assumption before you trust them.
- “Up to” is doing work. A forwarding rate prefixed with “up to” usually means a best case under ideal conditions. The number you care about is the sustained, non-blocking rate.
- Latency and buffer depth are the third and fourth numbers the headline pair leaves out. A switch can be non-blocking and still add unacceptable latency or drop bursts because its buffers are shallow. Those live further down the page, and for some roles they matter more than either headline figure.
The short version: do the two pieces of arithmetic yourself. Sum the ports for bandwidth, multiply gigabits by 1.488 for packets. A datasheet that meets both of your computed numbers is telling you the truth. One that meets only the flattering one is telling you a number, and leaving you to discover the rest in production.