On Fri, 16 Nov 2001, Dag Brattli wrote: > I'm not so sure about the mtt theory since the stack will always use a > bit of time before sending the next frame (protocol processing), so there > will always be a little delay before sending the first frame. Note that
Hi Dag, seems you were right, the latency was probably due to FAST_RR not enabled. However, would you like to comment on this output: irlap0 state: LAP_XMIT_P device name: irda0, hardware name: VLSI-FIR @ 0x3000 caddr: 0x32, saddr: 0xf5a534d0, daddr: 0x56c9e2d7 win size: 3, win: 3, line capacity: 57600, bytes left: 57600 tx queue len: 0 win queue len: 0 rbusy: FALSE mbusy: FALSE retrans: 0 vs: 0 vr: 7 va: 255 qos bps maxtt dsize winsize addbofs mintt ldisc comp tx 1152000 500 256 3 0 0 12 rx 1152000 500 2048 7 0 1000 12 It was taken during a MIR connection between ob800 and a T39m. Do you think the mintt=0 for tx to the phone is a non-issue? Agreed, we need some time for protocol processing - but the boxes tend to be pretty fast these days and the in-kernel stack isn't that slow it would take order of ms for turning the connection. In fact irdadump reports typically 0.05-0.08 ms diff-time for i:rsp (pf=1) to i:cmd (or rr:cmd) even for the ob800 (P5-166). I haven't tried to follow all the paths but my guess is this is all done in the netif_rx() bottom half context. Userland stack might be different of course. My knowledge is certainly pretty limited, but IIRC most IrDA transceivers have transmitter receive latencies of about 1ms - even the fastest high-volume consumer-grade devices which I know do not guarantee less than 100 usec. Therefore, if this mintt=0 claim would be too optimistic (probably because it seems to behave with other implementations ;-) do you believe we are safe here not to run into trouble? Do you know any transceiver specified for zero tx-turn latency? Well, sub-usec latency might be possible with special semiconductor junction design, gated receiver and all this fancy stuff - but does such a thing exist and would it be used for mass market with GSM/GPRS bottleneck? Martin _______________________________________________ Linux-IrDA mailing list - [EMAIL PROTECTED] http://www.pasta.cs.UiT.No/mailman/listinfo/linux-irda
