The data bandwidth requirements of Wi-Fi devices have been increasing tenfold every five years while the number of Wi-Fi devices inside the home is doubling every year. This means the data processing requirement of wireless access points is growing at a compounded rate. With the advent of devices with 802.11ac connectivity, the trend is expected to continue at a much more rapid pace as we anticipate increases in data rates by 3x from about 450 Mbps for 802.11n 3×3 to 1.3 Gbps for 802.11ac.

With the ever increasing data rates and number of devices managed by access points (AP), hardware solutions for AP need to be scalable, not only with respect to these parameters, but also with respect to the services provided at the point of access. Quality of Service (QoS) and firewall requirements for enterprise APs are also two extremely important factors that need to be considered.

In addition, cloud-based management requires the AP to be a lot more intelligent and capable in handling data packets by making local decisions.

Ikanos Vx575 - MIPS interAptivIkanos designs MIPS-based processors for networking applications

This means the current generation of APs used in enterprise, outdoor or small cell applications need to be improved for higher throughput, number of stations, security, packet processing, and more.

An overview of a traditional networking architecture

Most of the AP wireless chipsets are developed for client side and retrofitted to APs with a host processor to perform all the functions required in the access point. This approach is not conducive to scaling and is not efficient in terms of power budget, number of security sessions, fast response times, U-APSD/LP latency, and execution of contention free protocols especially with data rates in the 1.3 Gbps range.

Power Budget

Most deployments of access points require them to be powered over Ethernet (PoE) to avoid additional power connections. Though some of the current versions of PoE provide very high power, most current PoE Ethernet switches (802.2af compliant Cat-5) have a power budget of 14.5W per device.

Increasing the number of cores and/or increasing the frequency of the host processor to process the packets exceeds these power budgets. In addition, the power consumption of the 802.11ac baseband and RF will be higher for superior data rates.

Scalability

The number of security sessions supported by wireless devices is usually between 32 and 64. This forces the centralized controller to handle the crypto and security functions for all the APs.

A significant portion of the performance is thus consumed in per packet processing (classification, packet editing, and actions like filtering) and with the higher packet/data rates of 802.11ac, the host processor will be a bottleneck.

1-1-Networking AP hardware architectureThe architecture of an Access Point (AP)

Quality of service

APs have to support multiple voice, video and data connections. A very sophisticated QoS is required to support these different services across multiple stations/users. Processor-based QoS algorithms are known to be sub-optimal in both performance and efficiency in supporting things like voice-video synchronization as they represent two different access categories with a large number of clients.

Since jitter and sensitivity are important for voice traffic, QoS should be implemented as an edge function for it to be effective. Adding QoS directly in the controller does not have the same effect as implementing at the edge, as there are a lot of buffering levels in the middle. QoS for short packets can essential only handle voice data.

Low power client latency

The client power consumption depends on the response time of the access point. In U-APSD and legacy low power modes, the client wakes up, sends a null data/PS-POLL packet and waits (awake) till it receives a packet. In the AP, the null data/PS-POLL packet is sent to the host; the host then processes the packet and queues it to the Wi-Fi client.

Most often, the packet for the low power device gets queued behind the already scheduled queues and gets backed up, introducing a lot of latency in the response. Note that most client (as well as AP) Wi-Fi chips have only 5-6 hardware queues for 4 ACs and 2 for management/control.

Spectrum analysis

Since Wi-Fi spectrum sits in the unlicensed band, the network is uncontrolled; this means the higher-level protocols need to know about the activity in the current channel and the other adjacent channels. The data collected will be useful for:

  • Channel selection
  • Scenario replay and diagnosis of packet drops
  • Network planning
  • Continuous updates to WMM parameters like CW per AC

Some of the current system solutions have an additional Wireless LAN card to address these requirements which is expensive and redundant.

Coming up in our next article

Stay tuned to our blog – in our next two articles in this networking-focused miniseries we will present a new framework for access points and the architecture to support it.

About the author: Narayanan Raman

Profile photo of Narayanan Raman