netdev
[Top] [All Lists]

TGe (QoS in wireless)

To: netdev@xxxxxxxxxxx
Subject: TGe (QoS in wireless)
From: Vladimir Kondratiev <vkondra@xxxxxxx>
Date: Wed, 9 Jun 2004 09:36:16 +0300
Sender: netdev-bounce@xxxxxxxxxxx
User-agent: KMail/1.6.2
I will give short overview for TGe.

TGe is working group in IEEE, who is working on QoS for 802.11 wireless 
network. This work is close to completion, final draft will be voted and 
become standard shortly.

Next generation of wireless cards will support TGe.

Now, to the business.

Single radio channel used.

There are 2 access disciplines: 

- EDCA, which is CSCA, or contention based. 4 AC (access categories) exist, 
they differ by channel access parameters. Parameters include backoff, 
contention window and TXOP (more on this later). Parameters are set by AP and 
may be changed during operation.

- HCCA, which is contention free access. AP (access point) gives TXOP directly 
to each STA, deciding who will speak.

TXOP stands for transmitt opportunity. When each AC gets access to the 
channel, it is allowed to do burst (without backoff) for TXOP period, which 
is several ms.

Traffic classified into 16 categories, or TID. Each packet get its TID.
TID 0-7 get mapped statically to 4 AC, roughly it corresponds to TOS bits.

TID 8-15 mapped to TSPECs. TSPEC is traffic specification, which consist of 
bandwidth specification (throughput, delay etc. - see RSVP for example) and 
TCLAS - classifiers passed to AP side to classify packets.

TSPEC may be for HCCA or EDCA access. In case of HCCA, it get simple periodic 
schedule, set by AP. AP will poll STA accordingly to this schedule, and Tx to 
STA accordingly to schedule. In case of EDCA, it is just bandwidth allocation 
control.

Admission. AP may enforce admission control for some (actually, 2 high prio) 
AC, prohibiting its normal operation and requiring TSPEC establishment in 
order to use these categories.

I did not touched B-ACK (block ack), local broadcast, and DLP (direct link 
protocol) that deals nothing with QoS.

----------below is my interpretation-----------
This leads to the following channel model :
- single Rx channel
- 4 FIFO/DMA rings, one per AC.
- 1 to 8 FIFO/DMA rings for TSPEC traffic.
 In-NIC scheduler decides which FIFO get to the air. Time for decision is 
about 10 us, thus it can't be done on host.
Decision is the following:
- poll from AP -> TSPEC
- channel is clear, random backoff expired -> AC

Vladimir.

<Prev in Thread] Current Thread [Next in Thread>