I haven't actually looked at the hermes code in quite a while, but
I did write a device driver for the Aironet 655 ISA card (about 5
years ago), so I think I might have an idea of where your assumptions
are going wrong (although I am only guessing).

The Tx interrupt you are referring to is almost certainly generated
once the packet has been successfully buffered for transmission on
the PCMCIA card, as opposed to actually transmitted. At this point,
the PCMCIA card is informing the host system that it has the packet
and that the host system can now free the buffer etc. and start to
send in the next packet, which will be queued for transmission. This
process looks to take about 100us (for a 1300 byte packet) from your
timing.

The 600us you see when you jam packets through as fast as possible
is most likely the delay caused by the card completing the transmission
of the previous packet, freeing its own internal buffers and then
interrupting to indicate that there is now space for another packet
to be loaded in.

If the data flow wasn't done this way, you would see much lower
throughput through your wireless link.

Cheers,

Bob Edwards.

Steve J wrote:
>
> Hi,
> I am dong some experiments and I was just trying to
> figure out the actual transmission time.
>
> SO I call dogettimeofday() just before the actual
> transmission is triggered in dldwd_xmit()
>
> i.e. before - "err = hermes_docmd_wait(hw,
> HERMES_CMD_TX | HERMES_CMD_RECL, txfid, &resp);"
>
> and again call the dogettimefday() whe I receive an
> interrupt from the card abt the succcessful
> transmission.
>
> I measure the interval betwen these two calls as
> actaul trasnmission time.
>
> When I start a UDP application which sends 1300 bytes
> packet at interval of 10msec, the tx_time come out to
> be order of 100 microsecond. IT is beyond my
> understanding how this is possible as I set the card
> rate limit as 2Mbps. So given this setting the tx_time
> should come out as 1300*8/2*(10^6) = 5msec
>
> Also if this UDP allpicalltion sends packet as fast
> as posible, i.e.without any sleeep calls(), the the
> tx_time come out to be ~600 microsecond. This is still
> not correct and also why should the transmission time
> should change depending on the packet geneeration rate
> of application ??
> Kindly help
> -S
>
> __________________________________________________
> Do You Yahoo!?
> Yahoo! Movies - coverage of the 74th Academy Awards®
> http://movies.yahoo.com/

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]