3 ms·
Right. But the overhead difference between 48 and 64 for large-packet data is only about 2%, so data people were OK with 48. 32 would have been a bad choice fo
by tlb 8y ago
Right. But the overhead difference between 48 and 64 for large-packet data is only about 2%, so data people were OK with 48.
32 would have been a bad choice for data, as a large fraction of all internet data packets are 40-byte TCP ACKs.
- jzwinck 8y agoAren't TCP ACKs 20 bytes by themselves? The 40 byte total includes IP, which would not be needed if running on ATM, right?
- tlb 8y agoIn most uses of ATM, a single VC (virtual circuit) carries all the data between a pair of switches. The protocol to set up a new VC is pretty expensive, so it's impractical to set up a VC for each TCP connection or even between each pair of endpoint hosts. So mostly, VCs carry a stream of packets between switches, each with their own header.
- jzwinck 8y agoIn that case doesn't it mean there are 40 bytes of TCP/IP headers and 5 bytes of ATM headers plus a mere 8 bytes of payload per TCP packet?
- toast0 8y agoATM packets (cells?) are fixed size, but you can fragment higher level protocols across them. It's not common to run IP level fragmentation down to ATM sizes.