kenn at linux.ie
Sat Sep 9 23:30:59 IST 2000
On Sat, Sep 09, 2000 at 11:14:13PM +0100, Niall O Broin wrote:
> On Sat, Sep 09, 2000 at 04:42:44PM +0100, Kenn Humborg wrote:
> > Well, to be fair, this conversion should be happening on
> > a byte-by-byte basis, therefore the additional latency should
> > be of the order of a byte time, ie (1/11520) = 87uS.
> This only counts the serial transmisssion time and not whatever time is
> taken for processing in the TA and the PC. Also, the latency of 1 byte is of
> academic interest - the latency of a whole packet is what matters. For
> example, a ping with the default packet size of 64 bytes suffers from
> additional latency of 64 x 87 x 2 (there and back) =~ 1 ms. That doesn't
Nope. Because the conversion from 10-bit to 8-bit of the _second_ byte
gets done during the transmission of the _first_ byte. (Assuming that the
TA is acting as a straightforward modem, i.e. a physical-layer device,
and not on a packet-by-packet basis, like a router or bridge would.)
When a conversion device streams the data out at the same time as receiving
it, then the additional latency to do byte conversions is minimised.
Store-and-forward, like IP, is a different matter.
I deal with this stuff every day...
> sound bad, but there must be other factors involved because I know I've read
> on some game site that there's quite a difference in ping time between
> internal ISDN cards and external serial TAs - can't find a reference,
> unfortunately. Somebody else (Vin ?) who does actually use ISDN for gaming
> mentioned this earlier AFAIR.
Perhaps they act more like routers, then. Could the TA's serial interface
be talking a packet-based protocol like PPP? In that case, I agree, it's
a seriously broken design.
More information about the ILUG