- Know the definitions of latency (delay), throughput, and jitter.
- Know and understand the different types of delay.
- Understand the concept of effective throughput.
- Understand the empirical relationship between the delay and utilization of a data channel - the formula in 27.7.1.
- Know what QoS refers to, that it is defined in terms of constraints on delay, throughput, and jitter.
- Know what fine- and coarse-grain QoS are.
- Be able to discuss how QoS is implemented in routers - the four steps.
- Quantitative Measures of Networks
- How Technologies Can Provide Priority for Some Traffic
- Refer to Figure 27.1.
- Major measures of network performance:
- Latency (delay): time required for data to travel between two points on a network.
- Throughput (capacity): how many bits per second can a channel deliver?
- Jitter (variability): the variance of delay (how much change in delay?
- Refer to Figure 27.2 on page 472.
- Types of delay:
- Propagation Delay: delay in transmission medium
- Access Delay: access-to-medium delay - for example, due to contention for shared media.
- Switching Delay: time to forward a packet - usually not significant
- Queuing Delay: in switch or router, time waiting to be selected for transmission - accounts for most delay in the Internet
- Server Delay: elapsed time on server between arrival of request and transmission of first response - often much of what users perceive as Internet delay is really server delay.
- Maximum hardware data rates are rarely, if ever, achieved, because of overhead in the form of such things as headers, trailers, control information, limitations on window size, address and name resolution delay, connection set-up and shutdown, congestion, and packet loss.
- Effective Throughput is what users actually care about.
- Effective Throughput varies with the application in use.
- The terms bandwidth, speed and throughput are often used interchangeably. This is an unfortunate abuse of language.
- Throughput is a measure of capacity, not speed. It's a matter of how many bits per second of information can flow past a given point in a communication channel.
- Jitter is a measure of how much network delay varies with time.
- Suppose that an audio performance is sampled every 125 microseconds, the result placed in a packet, and transmitted to a receiver R.
- If there's no jitter in the channel, and if each packet can be played back as soon as it arrives at its destination, the playback rate will exactly match the sampling rate, and the playback will sound as it should. Note that it doesn't matter how large or small the delay is - only that the delay is constant.
- If everything is as above, except that there's significant jitter, then some packets won't be played back at the right times, relative to other packets. So the playback won't sound right.
- Through the use of real-time protocols it is possible for the receiver to compensate for some jitter, but only if the jitter does not become too great.
- 27.7.1 Utilization as an Estimate of Delay
- In practice a network operated at nearly maximum capacity will suffer from severe delay.
- Generally the amount of delay on a network can be estimated as:
D0 D = ----------- (1 - U)Where D0 is the delay when the network is idle, and where U is the fraction (between 0 and 1) of the maximum network throughput currently being utilized. - In other words, the delay on a network tends to be inversely proportional to the unused channel capacity.
- In still other words: the speed of bits or packets moving along a network tends to be proportionate to the unused channel capacity.
- Therefore, if we want to insure that a network will operate at close to minimum delay, we must insure that it does not operate very close to maximum utilization.
- Therefore, when throughput at maximum allowable utilization is not adequate, it's probably time to get technology with higher throughput - e.g. get a 100 Mbps Ethernet if operating the 10 Mbps Ethernet at 65% does not yield adequate throughput. Don't try to operate the 10Mps network at 95% of capacity.
- 27.7.2 Delay-Throughput Product
- The number, NB, of bits in a communication channel between
points A and B is:
NB = D X TWhere D is the delay between A and B, and T is the current rate of flow through the channel, in bits per second. - One way to think about the equation is that if the rate of flow is T bits per second, then it will take D seconds for a bit at position A to make the journey to B. At that time, all the bits that were between A and B will have flowed past B. The amount of the flow is D seconds X T bits/second - which is also NB - the number of bits that were in the channel between A and B.
- The number, NB, of bits in a communication channel between
points A and B is:
- To measure throughput and jitter is relatively simple.
- Measuring delay is more difficult because of the need to synchronize the clocks of the sender and receiver.
- Performance measurement in a network is also hampered by the rapidity with which traffic conditions can change, and by the fact that the measurement procedures themselves can affect performance.
- Aggregate data traffic is bursty and 'self-similar' - meaning that the statistical patterns remain the same in larger and larger aggregates of network data traffic.
- Passive measurement refers to monitoring without injecting additional traffic into the network. (Active monitoring does inject traffic.)
- For example, a router performs passive monitoring if it increments a counter each time it processes a packet.
- Passive monitoring tends not to greatly affect the performance it attempt to measure.
- There is an IETF standard called NetFlow for passive network measurement.
- Network Provisioning is designing a network to provide desired levels of service.
- Quality of Service (QoS) is defined in terms of constraints on delay, throughput and/or jitter.
- Refer to Figure 27.3.
- QoS may be provided in a fine-grain manner. For example, each TCP connection might have its own QoS requirements.
- QoS may be provided in a course-grained manner. For example a
client might have to pick one broad class of service that all its
traffic must use.
- 27.11.1 Fine-Grain QoS and Flows
- Refer to Figure 27.4.
- Telephone companies that did early work on QoS felt that fine-grain provisioning would be needed to support voice communication on a packet network.
- It turned out that routers don't have the processing power to implement per-flow QoS, and users have a hard time figuring out what to ask for anyway, so the fine-grain approach to QoS has little prominence nowadays.
- 27.11.2 Coarse-Grain QoS and Classes of Service
- With the coarse-grain approach, there are classes of traffic, and each class has certain QoS characteristics.
- Refer to Figure 27.5.
- The Four Steps of QoS Implementation (performed by a router or
switch)
- Classification and Policing: Router assigns flow identifier to arriving packet, drops packets if customer sends faster than agreed.
- Forwarding Computation: This takes the flow identifier into consideration.
- Output Queuing: The idea is to have separate queues for separate QoS requirements.
- Traffic Scheduling: The scheduler services the queues according to a scheme that assures the QoS requirements are met.
- Figure 27.6 illustrates some of the many traffic scheduling algorithms.
- Technologies and protocols related to QoS:
- RSVP and COPS
- DiffServ
- MPLS
- RSVP is the Resource ReSerVation Protocol and COPS is the Common Open Policy Services protocol. RSVP relates to fine-grain QoS - individual TCP or UDP sessions, hence is seldom used. COPS is a companion protocol to RSVP for policing.
- The IETF created Differentiated Services (DiffServ) to define coarse grain QoS, through the use of the TYPE OF SERVICE fields in IP headers. DiffServ has not met with great acceptance.
- MPLS is MultiProtocol Label Switching. It involves using encapsulation with an MPLS header (tunneling) to route certain classes of data, for example voice data, along a special path provided by a series of MPLS-capable routers.