------------------------------------------------------------------------------ CSC358 Tutorial 5 Notes Below are some notes and sample solutions to the tutorial questions. These notes are only meant to provide the necessary amount of information for you to verify your own work and to help you recall the discussions in the tutorial. Simply reading these solutions does NOT convey the same learning experience as attending a tutorial (not even close). The tutorial materials are a mandatory and important component of this course, so don't miss any of them! ------------------------------------------------------------------------------ Question 1 Discussed in the tutorial. Review the lecture slides. ------------------------------------------------------------------------------ Question 2 EstimatedRTT_n = alpha * sum_{i=1 to n-1} (1-alpha)^(i-1) S_i + (1-alpha)^(n-1) S_n When n -> infinity, the above becomes (alpha/(1-alpha)) * sum_{i=1 to inf} (1-alpha)^i S_i ------------------------------------------------------------------------------ Question 3 As shown in the tutorial, AIAD does not converge to fairness. If the two connections decrease with the same slope, the bandwidth allocation of the two connections just oscillates back and forth. If the one connection decreases with double the slope of the other, the allocation converges to allocating all bandwidth to the connection that decreases more slowly. ------------------------------------------------------------------------------ Question 4 (a) The loss probability is 1 over the number of segments sent in one increase-decrease cycle. This number is W/2 + (W/2 + 1) + (W/2 + 2) + ... + W (b) Ignore the (3/4)W in the result of Part (a) since W^2 >> W. Find W in terms of L. Then the average throughput is (3/4) * W * MSS / RTT (the factor 3/4 here is because it is, as discussed in the lecture, the mid-point of 1/2 and 1). (c) To achieve 10 Mbps, the loss rate must be at most 2e-4. To achieve 10 Gbps, the lose rate must be at most 2e-10. This is very hard to achieve in reality, therefore the current TCP is not suitable for super high throughput communication. New protocols are desirable.