[Ns-developers] questions and patch
Mathieu.Lacage at sophia.inria.fr
Tue Jan 17 22:17:30 PST 2006
Francesco Gringoli wrote:
> I must admit the DRAFT is not very plain about the time between two
> consecutive TXOPs in CFP if CFPOLL+ACK is not supported. Starting
> from the phrase before the last one in clause 18.104.22.168.2 "CAP
> generation" I think you are right.
> "The HC may sense the channel and reclaim the channel after a
> duration of PIFS after the TXOP, if the channel remains idle". Well I
> would say "reclaim" could not be interpreted as "switch to the next
> TXOP" but since I don't find another place where this is stated
> better, I will interpret "reclaim" in that way.
Maybe it woul be worth asking the 802.11 commitee for a clarification. I
don't care that much though.
> To summarize, here is the scenario I think should happen where N1 is the
>> number of packets in TXOP1:
>> TXOP1 transmission:
> Uhm, what is the first ACK? Do you mean (ACK+PKT1) is a single packet?
no, it is the ACK for CFPOLL1. If you don't send it, the QAP thinks the
TXOP is not used and switches to the next TXOP.
> I would remove the last SIFS and if N1 is the maximum number of
> "nominal" packet QSTA1 can send inside its TXOP, I will say that
> TXOP1 must contain [ CFPOLL+SIFS+(N1-1)*(PKT+SIFS+ACK+SIFS) +PKT_N1
> +SIFS+ACK ]
> Hence we can fix
> However (this is the key point) to calculate the CFP duration we have
> to add PIFS between every consecutive TXOPs plus a PIFS at the end of
> the CFP. Only if we calculate CFP duration in this way everything is
> ok. Imagine what happens if we don't count the PIFS: if we have a lot
> of backlogged traffic in all stations, every one will not send
> QOSNULL since there is no time. The last station however will never
> be polled because the m_txopTimer always added up PIFS to the TXOP
> and if the CFP=TXOP1+TXOP2+TXOP3+... all the added PIFSs have
> consumed time that should be granted to last station. Then it has no
> time to complete the TXOP inside the CFP.
yes, hence my suggestion to add PIFS somewhere :)
More information about the Ns-developers