Now that applications such as IP telephony, video conferencing and virtual desktop must be delivered to mobile...
devices as well as the desktop, a new kind of intelligence must be spread to components throughout the network for a new level of LAN QoS, including the edge.
Access switches that once simply forwarded packets must now support QoS capabilities that were previously required only in routers and core switches. Wireless access points (APs) must manage and allocate wireless bandwidth, giving high priority service to specific applications, while allocating remaining bandwidth among less important applications and users.
Despite the growth in smartphones and tablets, enterprises are continuing to provide wired connections to desktop phones, workstations, and docking stations. This means that QoS capabilities must support both wired and wireless environments simultaneously.
Addressing LAN QoS for wireless traffic
The IEEE and the Internet Engineering Task Force (IETF) have developed and standardized QoS techniques that operate at Layer 2, Layer 3 and on 802.11 networks. Additional techniques have been developed that determine which traffic is dropped when congestion occurs, or that provide a guaranteed data rate to other applications.
Three applications that require LAN QoS
While some applications such as large file transfers consume a great deal of bandwidth, other applications are trickier because they consume less bandwidth, but have other requirements that must be addressed using QoS techniques.
VoIP typically requires less than 100 kb to support a phone call, but has stringent latency requirements. Each voice packet originating at one end of a conversation must pass through the originating network, including switches, routers and firewalls. Once it crosses the Internet, it then enters the receiving network. The packet then makes the return trip through the same series of components. To meet the latency requirements, voice packets must be forwarded by each network component ahead of packets carrying lower priority data.
Video conferencing requires more bandwidth than VoIP, but has similar latency requirements.
Virtual desktop QoS requirements differ between server-based and client-based techniques. When applications execute on the server, the client workstation simply displays data. Keystrokes and mouse movement are relayed to the server and echoed back to the screen. Little data moves so bandwidth requirements are minimal, but any delay between the server and workstation is highly disconcerting to the user.
With client-based virtual desktop, the server downloads the OS and application to the workstation at the beginning of the session. The application executes on the client so keystrokes and mouse movements echo locally. While the OS and application downloads result in a large amount of data moving across the network, latency is not an issue.
IEEE 802.1p VLAN prioritization has long been implemented in most switches. Three bits in the Layer 2 Ethernet header are used to specify one of eight priority levels, with application software selecting the appropriate priority level. Unfortunately, the recent trend toward allowing employees to use their own smartphones and tablets has complicated the use of this feature.
Much of the software available for smartphones and tablets is not designed for the enterprise market. The operating systems for these devices support use of VLAN prioritization, but applications designed for the much larger consumer market have not taken advantage of this feature. In response, wireless product vendors have added intelligence to APs or wireless controllers to detect high priority traffic, primarily voice. The AP or controller selects the proper priority and modifies packets to the appropriate level.
With VLAN priority selected, APs can utilize IEEE 802.11e to prioritize wireless traffic. Wireless stations that have data to send wait until they detect that no other station is transmitting. When the transmission ends, stations must wait a period of time, and cannot begin transmitting immediately. 802.11e defines a mechanism that allows a station with high priority traffic a shorter wait time than other stations. This permits it to begin transmitting before other stations can gain control of the wireless medium.
Wireless equipment vendors have added more non-standard features to reduce congestion. Stations limited to IEEE 802.11g are confined to the 2.4GHz band, but 802.11n provides access at both the 2.4GHz and 5GHz bands. Some stations that are able to operate on the 5GHz band tend to connect on the 2.4GHz band. APs detect these stations and steer them to the 5GHz band, which is typically less crowded. APs can also conduct load balancing by shifting a station connected to the closest AP to an in-range, but more distant AP with available capacity.
Extending traffic shaping and more LAN QoS to the access network
VLAN prioritization operates at Layer 2. Differentiated Services (DIFFserv) is a layer 3 packet prioritization facility that can extend across the wide area network. RFC 2474 defines the DSCP field, a six bit field in the IP header. Unlike the IEEE 802.1p standard that defines the use of each VLAN priority level, the RFC does not define how the DSCP bits are to be used. Network implementers can configure switches and routers to utilize the six-bit field to define required levels. DIFFserv is supported in IPv6 as well as IPv4.
Facilities such as traffic shaping, rate limiting, and other methods to guarantee throughput levels have been extended to access routers. Network administrators can choose among algorithms that determine which packets to drop in the event of congestion. Such algorithms include Weighted Round Robin (WRR), Shaped Round Robin (SRR) and Weighted Random Early Detection (WRED). Vendors have added Committed Information Rate (CIR) to guarantee specific throughput levels to critical applications.
LAN QoS techniques offer network administrators a means to meet application needs. Voice traffic will be carried over high priority VLANs and DIFFserv settings. Applications such as client-based virtual desktop, which are not dependent on round-trip time but generate large amounts of data, will be allocated as lower priority and may be rate limited to avoid blocking higher priority traffic. As future application types are deployed, they will be matched to current QoS techniques or to other, yet to-be-developed strategies.
David B. Jacobs of The Jacobs Group has more than twenty years of networking industry experience. He has managed leading-edge software development projects and consulted to Fortune 500 companies as well as software start-ups.