Lecture 3: 3rd October 2019

Question 1

What is scheduling, in general?

Answer 1

Distributing requests to access a resource in a manner determined by a system or algorithm.

Question 2

What is the scheduled resource in routers?

Answer 2

A place in the output queue, to gain bandwidth/capacity.

Question 3

What is the service of routers?

Answer 3

packet forwarding

Question 4

What are the service requests of routers?

Answer 4

packets arriving on input lines

Question 5

How and why are packets classified in routers?

Answer 5

inbound packets and control signals from the forwarding and routing policies are fed into the internal packet classifier. This is because there are multiple output buffers so packets will be sent to the appropriate ones by some measure or distinction.

Question 6

What do the switching fabrics of routers do?

Answer 6

Use forwarding/routing tables to move received packets from the packet classifier to output buffers.

Question 7

How is it decided which output queue is able to output in each turn?

Answer 7

The scheduler decides this.

Question 8

What is FCFS aka?

Answer 8

FIFO

Question 9

How does FCFS work?

Answer 9

1st to arrive is 1st out; one queue preserving chronological order.

Question 10

What are work-conserving scheduling algorithms?

Answer 10

Scheduling algorithms in which each packet is serviced ASAP, and as long as there is traffic the output will never be idle. Like a roundabout vs traffic light crossroads: as long as cars in, cars going out.

Question 11

Is FCFS work-conserving? Why?

Answer 11

Yes because as long as there is input into the queue there will be output, and each packet is output as soon as it reaches the end of the queue.

Question 12

What is the difference between work-conserving and non-work-conserving algorithms?

Answer 12

Work-conserving scheduling algorithms are like a roundabout vs traffic light crossroads: as long as cars in, cars going out. A work-conserving scheduler is a scheduler that always tries to keep the scheduled resource/s busy if there are submitted jobs ready to be scheduled. It’s called work-conserving because it conserves the ability to constantly do work if possible.

Non-work conserving algorithms are harder to implement, usually in software rather than hardware, so are usually used toward the edge of networks. Token buckets are used to implement “speed limits”.

Question 13

What is the FCFS conservation law?

Answer 13

The product of the mean link utilisation and the delay due to the scheduler for each flow a scheduler is managing is equal to a constant which limits the maximum performance of a scheduler without hardware changes. The mean link utilisation is equal to the mean arrival rate divided by the mean service rate.

Question 14

What are the consequences of the FCFS conservation law?

Answer 14

The mean delay due to a scheduler is inversely proportional to its mean link utilisation.

Question 15

Can routers control the arrival rate?

Answer 15

No

Question 16

How can routers change their delay (the delay they add)?

Answer 16

Routers can’t change the mean arrival rate as this depends on users. To change the lin utilisation, and thereby the scheduling delay, therefore, you must change the service rate, since the constant and arrival rate cannot be changed.

Question 17

Why are non-work-conserving scheduling algorithms better for real-time, interactive applications?

Answer 17

They can change their service rates, allowing them to smooth out burts and equalise delays between users. Packets will only be serviced after becoming eligible. This reduces jitter, gives more predictable traffic downstream, and reduces the amount of buffer space needed, but increases end-to-end delay, which could worsen TCP throughput and is more complex to implement.

Question 18

What are some desired properties of scheduling algorithms?

Answer 18

easy to implement; fair; protect flows from mis/behaviour of others; give a guaranteed performance; control admissions - security, bandwidth, congestion

Question 19

What are FPGA?

Answer 19

Field Programmable Gate Arrays: integrated circuits that allow programmers to set the function of hardware.

Question 20

What is the significance of FPGAs with respect to scheduling in routers?

Answer 20

They allow the implementation of complex scheduling algorithms directly in hardware, making them faster, and offsetting the issue with non-work-scheduling or complex algorithms being too slow due to their complexity.

Question 21

What is max-min fairness?

Answer 21

A method of allocating the capacity of the resource being scheduled to the users wanting it. Each flow gets the minimum of the capacity it requested and its equitable share.

Question 22

How do you calculate the allocation of a resource to flows with max-min fairness?

Answer 22

Each capacity gets the minimum of the capacity it requested and the equitable share of the remaining total capacity. Sort flows by demands, smallest to largest. iterate through each and calculate its capacity. The capacity for each will be the minimum of its demand and (the total max of capacity being allocated, at the start - the amount of capacity already allocated) / the number of remaining flows.

Question 23

What are the consequences of max-min fairness?

Answer 23

Each flow either gets amount asked for or a fair share o what’s left. Flows get no more than they need. Since smallest first, incentivised to be honest and not ask for more than needed.

Question 24

What is the degree of aggregation of a scheduling algorithm?

Answer 24

How and how quickly routers are able to uniquely identify and distinguish flows. The extent to which flows are grouped together indistinguishably.

Question 25

What should scheduling algorithms consider?

Answer 25

The number of priority levels, whether to be work-conserving, the degree of aggregation, and if and how to perform intraqueue servicing.

Question 26

What is simple priority queueing?

Answer 26

When you have k queues, with a queue i + 1 having a greater priority than i and being serviced before it.

Question 27

What are the properties of simple priority queueing?

Answer 27

Simple to implement, little processing needed, but poor fairness since flows in low priority queues get starved.

Question 28

What is WRR scheduling?

Answer 28

WRR = weighted round-robin = each queue gets a number of turns proportionally to its assigned weight.

Question 29

What are the properties of WRR scheduling?

Answer 29

Unfairness can arise from different mean packet size / weight in each queue since packets sizes can be unpredictable. Usually fair for long-term flows, however.

Question 30

What is class-based queueing?

Answer 30

A mechanism for queuing in which queues are described as being nodes in a hierarchical tree. Each parent node must have a capacity greater than or equal to the sum of its children. Child nodes can exchange capacity if one over and one under.

Question 31

What are the properties of class-based queueing?

Answer 31

Only a queueing mechanism - still need scheduling on top. Most widely implemented algorithm; focuses on resource allocation; allows arbitrarily fine-grained queues, to the level of one flow.

Question 32

What is the difference between queueing mechanisms and scheduler applications?

Answer 32

Queueing mechanisms describe how queues are made, maintained, and operated on. Scheduling algorithms decide how a queue is picked to have a resource at each turn.