Introduction to Queuing
Cisco routers support a wide variety of queuing methodologies. Some have been around quite some time. Others are more modern, and can be quite complex and difficult to understand. Due to the complexity of more modern queuing technologies, it is difficult to describe and understand them without first understanding the queuing basics that lead up to the more complex methods.
This section describes the following three queuing methods that lead up to the complex queuing methods discussed later:
First-In, First-Out (FIFO) Queuing
Fair Queuing (FQ)
Weighted Fair Queuing (WFQ)
First-In, First-Out Queuing
FIFO queuing is the most basic of strategies. In essence, it is the first-come, first-served approach to data forwarding. In FIFO, packets are transmitted in the order in which they are received. Keep in mind that this process occurs on each interface in a router, not in the router as a whole.
On high-speed interfaces (greater that 2 Mbps), FIFO is the default queuing strategy on a router. Normally, such high-bandwidth interfaces do not have problems getting traffic out the door.
Figure 15-2 displays the basic model of FIFO. Notice that there are three different sizes of packets. One potential problem of FIFO is that the small packets must wait in line for the larger packets to get dispatched. In the figure, the smallest packet is actually ready to leave before the largest packet is finished arriving. However, because the largest packet started to arrive at the interface first, it gets to leave the interface first. This actually causes gaps between data on the wire, which decreases efficiency.
)
Figure
15-2 FIFO
FIFO is not really queuing; it is more along the lines of buffering. The packets are routed to the interface and stored (queued) in router memory until transmittal. The transmission order is based on the arrival order of the first bit of the packet, even though the last bit may be still far away. Essentially, the outbound packet buffer is selected as soon as its outbound interface is selected.
Fair Queuing
FIFO queuing does not offer any way to permit packets that are "ready" to be transmitted to leave before packets that are still "preparing" to be transmitted. As was demonstrated in Figure 15-2, large packets, based on arrival time, can clog an outbound interface because their first bit was first to arrive on the interface.
Fair Queuing is a methodology that allows packets that are ready to be transmitted to leave, even if they started to arrive after another packet. Note that FQ is not an option in Cisco routers, but understanding FQ will help you to understand WFQ.
Using the same example as before, the effects of FQ are shown in Figure 15-3. The same data flow is sent to the egress interface, only this time the smallest packets are allowed to leave first because they are ready to leave before the larger packet.
FQ allows smaller packets to "cut the line" in front of larger packets that are still in the process of arriving. This process solves the FIFO problem of gaps between packets on the wire caused by the blocking by the large packets.
)
Figure
15-3 Fair Queuing
Weighted Fair Queuing
As mentioned, FIFO is often not the ideal queuing method on low-bandwidth interfaces. Different data patterns may suffer in a FIFO environment. Consider Telnet and FTP competing for the same egress interface. With FIFO, the small Telnet packets must wait behind the large FTP packets. With FQ, the small Telnet packets are allowed to leave once each packet has completely arrived at the interface. When a large FTP packet is ready to go, it is dispatched. Then, while another large FTP packet is building in the buffer, multiple Telnet packets are sent.
However, FQ does not take into account any parameters stored within the packet, such as type of service (ToS). Some small packets, such as voice, should have a higher priority than other small packets, such as Telnet. If size were the only delimiter (FQ), then small voice packets would be considered the "same" as small Telnet packets. This could cause delay and jitter in the voice quality.
WFQ starts by sorting traffic that arrives on an egress interface into conversations or flows. The router determines what the actual flows are, and the administrator cannot influence this decision. Basically, the conversations are based on a hash (combination) of the source and destination IP addresses, protocol number, ports, MAC addresses, DLCI numbers, etc. Not all values may be used to determine any flow. The ToS is not used to determine flow.
The administrator can define the maximum number of flows possible. The router performs the flow selection. WFQ dispatches packets from the front of any given flow only. Thus, a packet in the middle of flow #2 cannot be dispatched until all the packets at the front of flow #2 are sent. In other words, each flow is handled in FIFO order.
WFQ differs from FQ because it uses the ToS bits that travel within each IP header. Remember that FQ looks at when a packet finished arriving (relative time) to determine when it actually is dispatched. Thus, the priority of the packet specified in the ToS bits becomes a "weight" when dispatching packets through an egress interface.
WFQ multiplies this relative time by a mathematical variation of the ToS to determine the new "when to dispatch" number. For this description, and to simplify the math, ToS 7 = multiplier 1, ToS 6 = multiplier 2, down through ToS 0 = multiplier 8. In reality, the multiplier numbers are much larger, but on a similar scale.
Figure 15-4 shows how the WFQ system works. Three packets have arrived on this egress interface. The router, configured for FQ on this interface, has determined that there are three different flows. The administrator cannot impact the flow selection process. The relative arrival time is shown below the queue.
)
Figure
15-4 Weighted Fair Queuing
For FIFO, the largest packet would be dispatched first, followed by the medium one, followed by the smallest. FQ corrects this by sending the smallest first, then the medium one, then the largest. But in this new example, the medium packet has a much higher priority (ToS = 5) than the small packet (ToS = 0). Thus, WFQ adjusts the dispatch accordingly.
Remember that all values shown here for the "multiplier" are adjusted for simple mathematical examples. Real numbers are much larger, but on a similar scale.
The large packet starts arriving at time 10, but finishes at time 17. With a ToS of 0, the multiplication factor is 8. Thus, 17 ∴ 8 = 136. The medium packet starts arriving at time 11, but finishes at time 15. Its ToS of 5 has a multiplication factor of 3. Thus 15 ∴ 3 = 45. And finally, the small packet starts arriving at time 12, however it finishes at time 14. ToS 0 = multiplier 8, thus 14 ∴ 8 = 112.
Table 15-2 takes all these potentially confusing numbers and arranges them logically.
Table 15-2 Weighted Fair Queuing
|
Packet |
Start |
Finish |
ToS |
Multiplier (based on ToS) |
Dispatch (finish ∴ multiplier) |
|
Small |
12 |
14 |
0 |
8 |
112 |
|
Medium |
11 |
15 |
5 |
3 |
45 |
|
Large |
10 |
17 |
0 |
8 |
136 |
So, the medium packet is dispatched first, followed by the small packet, followed by the large one. So it seems that WFQ solves the problems of getting small packets out first and ensuring that higher-priority packets get fair usage of the bandwidth. However, because the administrator cannot control the selection of the conversations (or flows), WFQ does have a few issues. Consider Figure 15-5.
)
Figure
15-5 WFQ #2
In this example, the third flow has two packets. However, the second packet is a high-priority packet (ToS = 5). It is quite possible to have packets of various ToS in a single flow. Remember that dynamic flow selection is not based on ToS.
The problem here is that the high-priority packet in flow #3 cannot be dispatched until after the large packet in front of it (same flow) leaves. Packets within a flow are handled FIFO. The WFQ algorithm only works with the first packets in each of the dynamically created flows. And as mentioned, the administrator has no control over how packets get sorted into the flows.
Thus, in the scenario shown, although it would be nice (and probably desired) to have the high-priority packets leave first, it is not the case. The high-priority packet in flow #3 is actually the last one out the door.