Voice over IP Online Course
Lesson 4 - Quality of Service
B. Bandwidth: Share or Restrict
Bandwidth: Share or Restrict
- Dealing with inadequate bandwidth
- Restrict access
- PVC (ATM, Frame Relay)
- Integrated Services (IntServ), RSVP
- Share access but give priority
- Differentiated Services (DiffServ)
- Queuing techniques
With unlimited bandwidth, any number of telephone calls as well as audio, video and data can make their way down a cable. When the bandwidth is limited, some hard decisions have to be made.
In the previous section, bandwidth was discussed as a factor in quality of service. There are two approaches to rationing bandwidth. In the restrict approach, access to the bandwidth is parceled out until all of it has been taken. Then no more sessions are allowed to initiate until a “slot” becomes available again. In the sharing approach, all sessions are allowed to initiate, however, certain traffic is given priority and transmitted promptly through intermediate points, such as routers. Non-priority traffic is held back at the routers and switches until there is space for them on the cable. At times of congestion, non-priority traffic can move very slowly and can even time out, at which point the session is lost and retransmission are required. Although this seems unfair to non-priority traffic, such as data, it can usually stand being sluggish. In any case, all traffic is accommodated in some fashion and none is turned away.Restricting access
ATM and Frame RelayMany routers have ATM or Frame Relay interfaces simply because these are common WAN technologies used by service providers. ATM and Frame Relay uses PVCs (Permanent Virtual Connections). With the PVC method, one end point must request a circuit to the other end. If one is available, both ATM and Frame Relay will set up the PVC and the two sides can communicate. If no circuit is available, the session is declined and no communication is possible. A similar effect can be seen when a busy signal is encountered trying to telephone overseas if no circuit is available.
ATM and Frame Relay naturally restrict access because of their design. The same cannot be said for a packet switching network.
Integrated Services (IntServ)
IntServ is the IETF’s answer to providing QoS on an IP network by restricting access to available bandwidth. IntServ was first defined by the IETF in 1997. RSVP is the primary mechanism used by IntServ to request QoS parameters for a data flow over an IP network.Resource Reservation Protocol (RSVP)
An RSVP host would request specific QoS parameters from the network, for example 16 kbps, 100 ms stable delay, etc. RSVP routers would provide the parameters when they set up the session. A session, or end to end connection is called a data flow in RSVP speak. A particular bandwidth could support only so may data flows. For example, a 1.544 Mbps link could support 24 64 kbps data flows and no more. Once the bandwidth was used up, RSVP would decline any further requests until a slot was freed up. This type of system requires that both end-point hosts as well as intermediate routers support RSVP.RSVP places a heavy burden on routers since they must track each data flow to make sure that service guarantees are met. This then is the Achilles heal of RSVP. It cannot scale to the extremely large networks we can expect to see with VoIP in the future.
Differentiated Services (DiffServ)
DiffServ was defined by the IETF in 1999. DiffServ provides an alternative to IntServ for providing QoS on a network. It is designed to answer the weaknesses of IntServ.DiffServ flags individual IP packets with a priority level using the ToS field in the IP packet. Therefore routers can handle packets according to their class of service and need not track them. In other words, routers can handle packets individually as they normally do. They don’t need to track the data flow. In addition, the end point hosts don’t necessarily need to understand DiffServ; intermediate routers can flag packets appropriately instead. This approach scales well as systems grow larger.
Queuing
Once a packet that has been flagged with a class of service reaches a router or switch, how is it suppose to handle it? This is a relevant question since packets of different priorities may reach the router or switch at the same time and multiple packets of the same priority may also reach it. Many queuing methods have been devised of which the following is a small sample.- Class-Based Queuing (CBQ) — CBQ divides all user traffic into categories and assigns bandwidth to each class.
- Per-VC Queuing — Per-VC queuing uses a separate output buffer for each virtual circuit. Each buffer can be given a priority, so voice virtual circuits (for example) can be given precedence over other data-carrying virtual circuits.
- Random Early Discard (RED) — This method relies on rules based on probability to instruct a router to begin dropping packets when established queuing thresholds are crossed.
- Weighted Fair Queuing (WFQ) — Each packet stream that WFQ is applied to is buffered separately and receives bandwidth on a variable, weighted basis.
- Weighted Random Early Discard (WRED) — WRED tries to identify the low-priority traffic and randomly discard those packets when congestion occurs
[Top of page][On to next section]
