Do You Hear What I Hear?�Part IX: Queuing Solutions for QoS
Concluding our series on VoIP Quality of Service (QoS), our previous installments have looked at some of the key factors surrounding the quality of the voice connection:
Part I: Defining QoS
Part II: Key Transmission Impairments
Part III: Dealing with Latency
Part IV: Measuring "Toll Quality"
Part V: Integrated Services
Part VI: Resource Reservation Protocol
Part VII: Differentiated Services
Part VIII: Multiprotocol Label Switching
In several recent tutorials, we examined the Integrated Services (intserv) and Differentiated Services (diffserv) projects of the Internet Engineering Task Force (IETF) that are designed to provide Quality of Service (QoS) capabilities for VoIP and multimedia networks. We also looked at Multiprotocol Label Switching (MPLS), which started as several vendor-proprietary solutions (and in those early days called tag switching), which were later melded into an IETF-supported standards development. To conclude our look at QoS, we will briefly define some of the queuing solutions that are currently being pitched in the router marketplace alongside those of the above standards.
To begin, a queue is a waiting linesomething that you have undoubted experienced at a toll booth on the highway, at the grocery store, or at a movie theater. Packets that are on their way from a source to a destination experience queues as well, as they wait their turn to be processed at a router, or enter a stream of data that must make a speed change, such as when a high speed LAN (such as fast Ethernet operating at 100 Mbps) interfaces with a lower speed WAN (such as a T-1 line operating at 1.544 Mbps). The simplest form of queuing is called First In First Out, or FIFO. As the name implies, the arrival order of the element (be that a car, human or packet) determines what gets the first available service. FIFO queuing does not have any mechanism to prioritize one traffic flow above another, nor does it have a way to allocate at least some of the available resources to each partywhoever gets their first grabs the first available resource or bandwidth.
Several improvements on FIFO queuing have been developed that are often used to improve QoS within router-based internetworks:
- Weighted Fair Queuing (WFQ): defines a resource allocation scheme based upon some type of differentiation in data flows. Each of the flows or traffic classes is assigned a mathematical weight, and then an algorithm allocates the available resources according to that assigned weight. Thus, low volume traffic flows can be given preference over higher volume traffic flows, which would otherwise consume an inordinate share of the network resource.
- Random Early Detection (RED): is sometimes called a congestion avoidance scheme, as it uses a proactive approach to manage the traffic queues before they fill up. RED makes use of the flow control mechanism built into the Transmission Control Protocol (TCP), that cause a sender to decrease its transmission rate when it learns that packets in that data stream have been dropped because of network congestion. If the queue length increases, additional packets are dropped. This sends an additional signal to the sender to further decrease the rate at which packets enter the network. As packets are dropped, the queue size decreases, which in turn decreases the queuing delay, and thus stabilizes network operation before a more serious problem occurs.
- Weighted Random Early Detection (WRED): an enhancement to RED which drops packets based upon a priority scheme. Different traffic flows are identified, and assigned values within the Precedence field of the Internet Protocol (IP) header. When network congestion occurs, packets with higher priorities are less likely to be dropped than those with lower priorities. This allows the network to maintain stable queue lengths, while giving priority to certain traffic flows that have require a higher QoS.
WFQ, RED and WRED are typically implemented within router-based multimedia networks, and therefore a perusal of your favorite router vendor's website should yield some additional technical and implementation details for further research.
Copyright Acknowledgement: © 2005 DigiNet ® Corporation, All Rights Reserved
Mark A. Miller, P.E. is President of DigiNet ® Corporation, a Denver-based consulting engineering firm. He is the author of many books on networking technologies, including Voice over IP Technologies, and Internet Technologies Handbook, both published by John Wiley & Sons.