Ethernet uses an exponential backoff to delay before resending.
To be precise, let c denote the number of attempts to send a packet that
have resulted in a collision. After collision c, we pick a
(randomly distributed) number, t, in the interval [0, T * 2^c),
where T is the initial period to wait. We then wait time t and resend.
If the packet again collides, c is incremented, thus doubling the interval
from which t is chosen. If c becomes too great, no further attempts are
made to resend the packet, and it's dropped.
Partial credit: +2 for describing time delay before resending +3 for noting time delay is random +3 for noting the potential time delay is increasing +1 for noting backoff is exponential
A series of links connect the machines. Later links may have smaller
maximum transfer units (MTUs) than the initial link, and packets that
exceed their maximum packet size must be split into smaller packets, which
are reassembled at the destination.
Partial credit: +4 if confused MTUs for bandwidth
Necause the virtual memory system always gives equal portions of
physical memory to each process, a local replacement policy is necessary.
In the steady state, where the number of processes is not changing, local
replacement ensures that physical pages are not taken from other processes,
so each process maintains its fair share. Global replacement does the
Partial credit: +3 if said global replacement, but demonstrated some conceptual understanding +9 if said global replacement, demonstrated a good conceptual understanding, and justified by saying that when the number of processes increases, physical memory must be taken away from some processes to be given to a new process. +5 to +8 for saying local replacement but having a poor justifaction