A-MPDU transmission in net80211 on FreeBSD 8

[Sam Leffler]

That's great to hear.  I take it this is HT20?  I believe you can get 
40-60 Mb/s @ MCS15 w/o AMPDU and 80-100 Mb/s w/ AMPDU.  In HT40 I've 
gotten 160-180 Mb/s for TCP netperf (5GHz) and 220+ Mb/s for 
unidirectional streams (netperf -t UDP_STREAM).  Note that once you get 
to these higher rates things like TCP window sizes become important and 
doing things like TSO in s/w can give noticeable speed boosts.

	Sam

Alexander Egorenkov wrote:
> Hi,
> 
> thanks for your help, I implemented A-MPDU Tx in my Ralink driver now 
> and it works very good.
> I have average Tx rate about 4.5~5 MBytes/s now :-) The Ralink chip that 
> i have implements the frame queue in hardware so i had only to assign 
> sequence numbers, set BA window size and MPDU density and it worked, 
> lucky me :-)
> 
> Alex.
> 
> On Sun, Jan 31, 2010 at 8:20 PM, Sam Leffler <sam at errno.com 
> <mailto:sam at errno.com>> wrote:
> 
>     Alexander Egorenkov wrote:
> 
>         Why doesn't 802.11 stack assign sequence numbers to A-MPDU frames ?
> 
> 
>     Because if net80211 does the assignment it may be wrong.  As the
>     comment says, if tx aggregation causes frames to be q'd above the
>     h/w then by the time they are sent OTA the pre-assigned seq# may be
>     invalidated by other frames going out ahead of it.
> 
> 
>         When sequence numbers are not assigned to A-MPDU frames, then BA
>         doesn't work with my AP.
>         I tried to assign sequence numbers to A-MPDU frames in my device
>         driver and then BAs worked
>         with my AP.
> 
> 
>     That is what the comment says to do.
> 
> 
>         And what is meant by aggregation queue ? Where is that queue anf
>         how do i use it ?
> 
> 
>     The aggregation q is the mechanism used to hold frames waiting for
>     additional frames to aggregated into an A-MSDU/A-MPDU.  The queue is
>     typically wherever the aggregation work is done.  Some devices do
>     this in h/w, others require the host handle this.  When done in the
>     host it can be done in the driver or above.  The intent has always
>     been to have net80211 implement tx aggregation that a driver can
>     fallback on but I never did the work.  All the 11n drivers I've done
>     have either handled tx aggregation in h/w or in the driver.
> 
> 
>         Thanks.
> 
> 
>         On Sun, Jan 31, 2010 at 4:43 AM, Sam Leffler <sam at errno.com
>         <mailto:sam at errno.com> <mailto:sam at errno.com
>         <mailto:sam at errno.com>>> wrote:
> 
>            Alexander Egorenkov wrote:
> 
>                Sorry, i posted the wrong comment.
>                Here is the comment which i don't understand:
> 
>                /*
>                            * NB: don't assign a sequence # to potential
>                            * aggregates; we expect this happens at the
>                            * point the frame comes off any aggregation q
>                            * as otherwise we may introduce holes in the
>                            * BA sequence space and/or make window accouting
>                            * more difficult.
>                            *
>                            * XXX may want to control this with a driver
>                            * capability; this may also change when we pull
>                            * aggregation up into net80211
>                 */
> 
>                Thanks.
> 
> 
>            What is unclear?
> 
>                   Sam
> 
> 
> 
>                On Wed, Jan 27, 2010 at 8:04 PM, Alexander Egorenkov <
>                egorenar at googlemail.com <mailto:egorenar at googlemail.com>
>         <mailto:egorenar at googlemail.com
>         <mailto:egorenar at googlemail.com>>> wrote:
> 
>                    Hi,
> 
>                    i'm implementing a device driver for a 802.11n NIC under
>                    FreeBSD 8
>                    und experimented with A-MPDU transmission. I looked into
>                    net80211 code
>                    and there is some code which implements this feature
>         but it
>                    worked not very
>                    well for me.
>                    I noticed e.g. that sequence numbers are not assigned to
>                    A-MPDU frames
>                    and found this comment in file ieee80211_output.c :
> 
> 
>                    /*
>                            * Check if A-MPDU tx aggregation is setup or
>         if we
>                            * should try to enable it.  The sta must be
>         associated
>                            * with HT and A-MPDU enabled for use.  When
>         the policy
>                            * routine decides we should enable A-MPDU we
>         issue an
>                            * ADDBA request and wait for a reply.  The
>         frame being
>                            * encapsulated will go out w/o using A-MPDU,
>         or possibly
>                            * it might be collected by the driver and
>                    held/retransmit.
>                            * The default ic_ampdu_enable routine handles
>         staggering
>                            * ADDBA requests in case the receiver NAK's
>         us or we are
>                            * otherwise unable to establish a BA stream.
>                     */
> 
>                    Can somebody elaborate this description to me please.
> 
>                    Thanks.
> 
>                    ALex.
> 
> 
>                _______________________________________________
>                freebsd-net at freebsd.org <mailto:freebsd-net at freebsd.org>
>         <mailto:freebsd-net at freebsd.org
>         <mailto:freebsd-net at freebsd.org>> mailing
> 
>                list
>                http://lists.freebsd.org/mailman/listinfo/freebsd-net
>                To unsubscribe, send any mail to
>                "freebsd-net-unsubscribe at freebsd.org
>         <mailto:freebsd-net-unsubscribe at freebsd.org>
>                <mailto:freebsd-net-unsubscribe at freebsd.org
>         <mailto:freebsd-net-unsubscribe at freebsd.org>>"
> 
> 
> 
> 
> 
>