26 Jul
2018
26 Jul
'18
8:03 p.m.
Clemens Ladisch <clemens(a)ladisch.de> writes:
David Kastrup wrote:
If we are aggregating 4 4-port interfaces used in two directions (USB
rates are half-duplex), we need 32 times the bandwidth. So we are
already at 1Mbps of raw data rate without any packaging. 90% of the
full-speed bandwidth can be reserved for isochronous transfers which
MIDI isn't.
Queuing theory tells us that independent events close to filling up a
given bandwidth tend to stack up, so in what order is a single
transaction translator going to schedule transfers? What will be the
effect on jitter? And events created by arrangers are not actually
independent: time codes and chord notes and drums are coming out
back-to-back.
You'll be getting to "nobody will ever need more than 640kByte"
territory sooner than you'll think...
--
David Kastrup
If you want to hook up several ones of those via
a USB hub, make sure
that this hub has a separate "Transaction translator" per port so that
a USB1.1 transaction from one Midi interface does not keep other
transactions from happening: the 12Mbps limitation should only be per
Midi interface, not per aggregating hub which can talk at 480Mbps to
the computer.
MIDI's speed of 0.03 Mbps is less than one percent of the USB 1.1 limit;
the lack of a transaction translator does not matter at all (unless
other, high-speed USB devices are connected to the hub).