Cisco Cisco Aironet 3700i Access Point 백서
© 2015 Cisco and/or its affiliates. All rights reserved. This document is Cisco Public Information.
Page 12 of 23
In conclusion, Cisco Optimized Roaming helps prevent a negative experience for Wi-Fi users. It monitors the
connection quality of all devices and proactively prompts poorly performing client devices to seek a better
connection much sooner than the client would do on its own. This is why Optimized Roaming is an essential
component of Cisco HDX.
Features of the Cisco High Density Experience with Turbo Performance
In a high-density environment, an ever-increasing number of clients per access point and higher speeds per client
bring associated scaling challenges in traffic (or packet) forwarding. After all, if one can provide very high speed
connectivity to clients but cannot provide the corresponding very high throughput between the wireless and wired
LANs (in both directions), the benefits of technologies such as 802.11ac and ClientLink 3.0 cannot be fully realized.
For this reason, Cisco is introducing a highly efficient and scalable feature, Turbo Performance, targeted
specifically to overcome scaling issues with high density. This feature is difficult for competitors to match.
Turbo Performance comprises algorithms for packet queuing, scheduling, and forwarding that scale effectively and
efficiently to a high number of active clients. As you will see, Cisco has measured little to no performance
degradation with up to 60 client devices simultaneously ramping up and consuming multimedia traffic, whereas
comparable competitive devices begin to show noticeable degradation at 20 devices for the same traffic patterns.
Why Turbo Performance?
Although 802.11ac gets a lot of attention for providing Gigabit-class performance, we need to look beyond the
airlink to what the access point needs to do in order to get traffic to and from clients as efficiently as possible.
We need to consider throughput in terms of packets per second (PPS). In other words, the actual payload size
(in number of bytes) can vary widely depending upon the applications in use. Two devices could be connected at
identical airlink rates and yet have different behaviors regarding throughput and packet processing since they are
running different applications.
As an example, with 802.11n, an access point might have had to forward 30,000 1500-byte packets per second
through the access point. But with 802.11ac, that number can easily approach 75,000 (or more) PPS. More PPS
means more load on the access point’s network processor. Correspondingly, to meaningfully keep up with the
means more load on the access point’s network processor. Correspondingly, to meaningfully keep up with the
demands of 802.11ac and high-density deployments, a better, fresher approach was needed.
When designing the Cisco Aironet 2700 and 3700 Series Access Points for 802.11ac and high density, Cisco
fundamentally rearchitected the data-forwarding mechanism. Instead of a traditional packet-based data plane and
scheduler, Cisco implemented a client-based data plane and scheduler. In essence, the traffic forwarding is now
handled by a lightweight but highly efficient packet scheduler.
Furthermore, both the Aironet 2700 and 3700 Series use a custom, purpose-built 802.11ac chip in which a
generous amount of DDR3 RAM is colocated and dedicated to each radio subsystem. The advantages here are to
enable a large number of buffers localized to each radio and provide a highly optimized lower MAC data plane for
so-called Aggregated MAC Protocol Data Units (A-MPDUs). The benefits are much less intensive processing for
the network processor and much more efficient packet scheduling, thereby delivering packet forwarding
(throughput) commensurate to 802.11ac speeds while doing so at a much greater scale than competitive products.
Figure 2 shows an access point with a traditional packet forwarding design compared to an access point with the
Turbo Performance packet forwarding design. In the traditional access point, the RAM is coupled only with the
network processor instead of with the radio subsystems.