5. Latency vs Throughput
5.3 What's the difference between bandwidth and throughput?
On the highway: Four lanes can carry, say, 8,000 cars an hour. That's the capacity. On a quiet Sunday only 500 cars use it. On a day with roadworks, maybe 3,000 get through even though everyone wants to. Capacity is what's possible; the count at the toll booth is what actually happened.
In the system:
- Bandwidth is the maximum rate a link or component can carry, like a 1 Gbps network card.
- Throughput is the rate you actually achieve. It's always at or below bandwidth, because of protocol overhead, retransmissions, slow senders or receivers, and other traffic.
A quick example: a 1 Gbps link sending pages of 2 MB each can deliver at most 1,000,000,000 ÷ 8 ÷ 2,000,000 = 62.5 pages a second, however fast your server is. (Remember Chapter 2: network in bits, storage in bytes, times eight.)
And bandwidth isn't latency: a link can carry huge amounts per second and still have a long round trip, like a wide highway between two far-apart cities.
Memory code: "Lanes are what's possible; the toll count is what happened." Bandwidth is the ceiling, throughput is the reading.