PROJECTENGRAM

END TO END THROUGHPUT AND BOTTLENECK LINKS

End to End Throughput and Bottleneck Links

Tags: #◇CSNETWK #◇Computer Science #◇Computer Networks #◇Networking Protocols #◇Graph Theory Basics


◇Overview

Definition: Throughput

◇Key Concepts

Topology Server Links ( Client Links ( Shared Middle Link ( Fair Sharing

◇Details / Theory

◇1. Concept Overview: Throughput & Bottlenecks

Definition: Throughput

Throughput is the actual rate at which data can be transferred from a sender to a receiver. It is determined by the "slowest" link in the path.

Definition: The Bottleneck Link

In any chain of links, the Bottleneck Link is the link with the minimum transmission rate. The entire path cannot be faster than this single link.

The Formula (Fair Share)

If NN connections share a common link with rate RR, each connection gets a "fair share" of R/NR/N.

Throughput=min⁡(Rs,Rc,RN)\text{Throughput} = \min(R_s, R_c, \frac{R}{N})
RsR_s: Rate of Server Link RcR_c: Rate of Client Link RR: Rate of Shared Link NN: Number of concurrent flows

◇2. The Network Topology

Scenario Data
!◇Pasted image 20251121094535.png
Topology: 4 Servers sending to 4 Clients (4 pairs). Server Links (RsR_s): 10 Mbps each. Client Links (RcR_c): 90 Mbps each. Shared Middle Link (RR): 400 Mbps total. Fair Sharing: The middle link divides its capacity equally among the 4 pairs.

◇3. Problem Analysis

Question 1

What is the maximum achievable end-end throughput (in Mbps) for each of four client-to-server pairs?

Analytical Breakdown:

  1. Identify Constraints: We need to find the minimum speed along the path for one single pair.

  2. Link 1 (Server Side): Rs=10R_s = 10 Mbps.

  3. Link 2 (Shared Middle): The total is 400 Mbps, but it is shared by 4 pairs.
    Per-pair capacity = 400/4=100400 / 4 = 100 Mbps.

  4. Link 3 (Client Side): Rc=90R_c = 90 Mbps.

  5. Calculate Minimum:
    min⁡(10,100,90)\min(10, 100, 90)

10<90<10010 < 90 < 100

  1. Conclusion: The server link is the slowest part of the chain.

Correct Answer: 10 Mbps

Question 2

Which link is the bottleneck link?

Analytical Breakdown:

  1. Compare Capacities:
    Rs=10R_s = 10 Mbps
    Shared Share (R/4R/4) = 100 Mbps
    Rc=90R_c = 90 Mbps

  2. Identify Lowest: The value 10 is the lowest.

  3. Map to Link: This value corresponds to RsR_s (Server Link).

Correct Answer: Rs

Question 3

Assuming that the servers are sending at the maximum rate possible, what are the link utilizations for the server links (Rs)?

Analytical Breakdown:

  1. Definition: Utilization=Actual Traffic FlowLink Capacity\text{Utilization} = \frac{\text{Actual Traffic Flow}}{\text{Link Capacity}}
  2. Actual Flow: The maximum throughput we calculated in Q1 is 10 Mbps.
  3. Link Capacity: The capacity of RsR_s is 10 Mbps.
  4. Calculation: 10/10=1.010 / 10 = 1.0 (or 100%).

Correct Answer: 1

Question 4

Assuming that the servers are sending at the maximum rate possible, what are the link utilizations for the client links (Rc)?

Analytical Breakdown:

  1. Actual Flow: The traffic arriving at the client is limited by the bottleneck, so it is still arriving at 10 Mbps.
  2. Link Capacity: The capacity of RcR_c is 90 Mbps.
  3. Calculation: 10/90≈0.111...10 / 90 \approx 0.111...

Correct Answer: 0.11

Question 5

Assuming that the servers are sending at the maximum rate possible, what is the link utilizations for the shared link (R)?

Analytical Breakdown:

  1. Actual Flow: There are 4 pairs, each sending 10 Mbps.
    Total Flow = 4×10=404 \times 10 = 40 Mbps.

  2. Link Capacity: The total capacity of the shared link RR is 400 Mbps.

  3. Calculation: 40/400=0.140 / 400 = 0.1.

  4. Alternative Method: Calculate per-pair utilization. Share = 10 Mbps. Capacity Share = 100 Mbps. 10/100=0.110/100 = 0.1.

Correct Answer: 0.1

◇Summary of Logic (Cheat Sheet)

ConceptFormulaLogic
Bottleneckmin⁡(Link1,Link2,...)\min(\text{Link}_1, \text{Link}_2, ...)A chain is only as strong as its weakest link.
Shared LinkCapacityeffective=RtotalN\text{Capacity}_{\text{effective}} = \frac{R_{\text{total}}}{N}Always divide shared bandwidth by the number of users first.
UtilizationThroughputCapacity\frac{\text{Throughput}}{\text{Capacity}}How "full" is the pipe? (0 to 1).

Critical Check:
If RsR_s is small (e.g., dial-up), the server is the bottleneck. If RR is small (e.g., congested backbone), the core is the bottleneck. If RcR_c is small (e.g., weak Wi-Fi), the client is the bottleneck.

◇Common Pitfalls

Forgetting that end-to-end throughput is determined by the bottleneck (slowest) link, not the fastest. Not dividing the shared backbone link RR by the number of active connections NN when multiple flows compete. Confusing link capacity with actual per-flow throughput — min⁡(Rs,Rc,R/N)\min(R_s, R_c, R/N) is the correct formula.

◇Summary

TL;DR — End-to-end throughput equals min⁡(Rs,Rc,R/N)\min(R_s, R_c, R/N) where RsR_s is server link rate, RcR_c is client link rate, and R/NR/N is the shared backbone divided among NN flows. The bottleneck link limits overall performance.

Related: - ◇Pasted image 20251121094535.png Source / Reference:

◇References & Navigation

CONFIRMATION // 理解度の確認

ACTIVE RECALL VERIFICATION

Select an answer to verify conceptual comprehension and reveal the technical rationale.

Q1

Suppose a server sends a large file to a client across a 2-link path with no other competing traffic: the server's access link has transmission rate RsR_s, and the client's access link has transmission rate RcR_c. What is the instantaneous end-to-end throughput?

Q2

A server has an uplink rate Rs=100 MbpsR_s = 100\text{ Mbps}, a client has a downlink rate Rc=15 MbpsR_c = 15\text{ Mbps}, and they communicate across a shared core backbone link of capacity R=1 GbpsR = 1\text{ Gbps} (1000 Mbps1000\text{ Mbps}) shared equally among N=10N = 10 simultaneous downloads. What is the per-connection end-to-end throughput, and which link is the bottleneck?

Q3

Using the same network from Question 2 (Rs=100 MbpsR_s = 100\text{ Mbps}, Rc=15 MbpsR_c = 15\text{ Mbps}, shared backbone R=1000 MbpsR = 1000\text{ Mbps}), suppose the number of simultaneous connections sharing the backbone increases to N=100N = 100. What is the new per-connection throughput, and where is the bottleneck now?

Q4

If a file of size F=30 MegabitsF = 30\text{ Megabits} (30×106 bits30 \times 10^6\text{ bits}) is transferred across an end-to-end path whose bottleneck throughput is 15 Mbps15\text{ Mbps} (15×106 bits/sec15 \times 10^6\text{ bits/sec}), approximately how long does the file transfer take (ignoring protocol header overhead and propagation delay)?

Q5

What is the fundamental difference between Transmission Delay (dtrans=L/Rd_{\text{trans}} = L / R) and Propagation Delay (dprop=d/sd_{\text{prop}} = d / s) on a single physical link?

End to End Throughput and Bottleneck Links | PROJECT ENGRAM