From mboxrd@z Thu Jan  1 00:00:00 1970
Message-ID: <11e7e2e36e9cd31ef7ed923da77a3224@hamnavoe.com>
To: 9fans@9fans.net
From: Richard Miller <9fans@hamnavoe.com>
Date: Tue,  6 Mar 2012 16:10:58 +0000
In-Reply-To: <2579b9.228186c7.D0Vx.mx@tumtum.plumbweb.net>
MIME-Version: 1.0
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Subject: Re: [9fans] usb slowness
Topicbox-Message-UUID: 66cd6da6-ead7-11e9-9d60-3106f5b1d025

> indeed, if i set the interrupt threshold control to 1 micro-frame (they
> default to 8), i get 10-11 sec dd, say, half as slow.

That seems a good change to make.  Particularly for usb mass storage,
which for some reason emulates a scsi transaction with three usb
transfers and therefore three interrupts per r/w operation (send
command, recv/send data, recv status).  To get anything like maximum
throughput, we'd also have to queue all three transfers at once so
that the usb controller could schedule them in adjacent microframes
(or maybe even the same one?).  Sending them separately amd
synchronously from the user-space driver will probably take long
enough to miss one or more microframes.

=C2=A1Hola! Nemo, should the interrupt threshold be set to 1 microframe
(125 =CE=BCs) to get better latency for all devices?  I can't see any goo=
d
reason for wanting to slow interrupts deliberately.