Explain why it is harder to develop a good timeout value for resending unacknowledged packets in TCP than it is at the data link layer?
When computing the timeout for TCP, more weight is normally given to past network performance than current. What affect do you think there would be if weighting were shifted the other way in favor of current value?