Date: Fri, 14 Dec 2001 14:37:52 +0100 From: Thomas Zenker <thz@Lennartz-electronic.de> To: Mike Silbersack <silby@silby.com> Cc: dillon@apollo.backplane.com, stable@freebsd.org Subject: Re: USB ethernet problem Message-ID: <20011214143752.A90727@mezcal.tue.le> In-Reply-To: <Pine.BSF.4.30.0112132157490.12948-100000@niwun.pair.com>; from silby@silby.com on Thu, Dec 13, 2001 at 10:05:23PM -0500 References: <Pine.BSF.4.30.0112132157490.12948-100000@niwun.pair.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, Dec 13, 2001 at 10:05:23PM -0500, Mike Silbersack wrote:
> I just noticed the thread with you guys talking about the tcp chagnes
> impacting the usb card. I have a better fix for you to test, thomas:
>
> Set the sysctl
>
> net.inet.tcp.local_slowstart_flightsize
>
> to 1.
>
> Then set sendspace back up to 32768 and rerun your tests.
>
> In short, the local slowstart flightsize is set by default to a huge
> value, meaning that tcp's slow start will be skipped, and you'll burst out
> 32K of data all at once. With it set to 1, slow start will run normally,
> and you won't overrun the card.
>
> Tell me if this change helps with your problem or not.
>
> Thanks,
>
> Mike "Silby" Silbersack
Ok, I tried this.
After doing some test, I decided, this can not be done with a quick
test, (because of contradictory results) so did a serie of tests.
It depends also on the transfer size (per connection). This is
obviously caused by the slowstart alghorithm.
in the end I collected a lot of data, this is now for a version
from July and the other from yesterday.
We should think about the defaults again, as there are people which
might install their laptops over USB/ethernet. I myself know now
what to do for my equipment, but people will have problems with
sysinstall.
here the test case:
host -- 100TX -- SWITCH -- 10TX -- NetGear EA101 -- USB -- client
data is transfered from host to USB/ethernet client with ttcp. The
short transfers correspond about the transfer size of distribution
parts for sysinstall (this is where I discovered this).
Manipulated was the net.inet.tcp.recvspace variable on the USB client,
net.inet.tcp.sendspace on the Host,
net.inet.tcp.local_softstart_flightsize on both at the same time.
Thomas
Host USB/ethernet both sides transfer transfer
sendspace recvspace local ssfls short long
(256K) (16M)
KB/s KB/s
+--------------------------------------------------------------+
| 4.3-STABLE 2001-07-17 |
+----------+--------------+-------------+----------+-----------+
| | 65536 | 65536 | 33 | 236 |
| 32768 | 65536 | 16 | 66 | 230-240 |
| | 65536 | 2-8 | 291 | 230-240 |
| | 65536 | 1 | 266 | 230-240 |
+----------+--------------+-------------+----------+-----------+
| 32768 | 16384 | 65536 | 303 | 268 |
| | 16384 | 2-... | 303 | 268 |
| | 16384 | 1 | 274 | 268 |
+----------+--------------+-------------+----------+-----------+
| 16384 | 16384 | 2-... | 280 | 265 |
+----------+--------------+-------------+----------+-----------+
| 65536 | 16384 | 2-... | 280 | 265 |
| | 16384 | 65536 | 355 | 268 |
| | 16384 | 2-... | 355 | 268 |
| | 16384 | 1 | 313 | 268 |
+----------+--------------+-------------+----------+-----------+
| 65536 | 8192 | ...-65536 | 355 | 268 |
| | 8192 | 2 | 357 | 268 |
| | 8192 | 1 | 313 | 268 |
+--------------------------------------------------------------+
| 4.4-STABLE 2001-12-13 |
+----------+--------------+-------------+----------+-----------+
| 32768 | 65536 | 65536 | 12 | 134 |
| | 65536 | 16 | 20 | 200-230 |
| | 65536 | 2-8 | 146 | 225 |
| | 65536 | 1 | 87 | 190 |
+----------+--------------+-------------+----------+-----------+
| 32768 | 16384 | 65536 | 318 | 267 |
| | 16384 | 2-... | 312 | 266 |
| | 16384 | 1 | 276 | 266 |
+----------+--------------+-------------+----------+-----------+
| 16384 | 16384 | 2 | 283 | 266 |
+----------+--------------+-------------+----------+-----------+
| 65536 | 16384 | 65536 | 371 | 268 |
| | 16384 | 2 | 361 | 268 |
| | 16384 | 1 | 320 | 268 |
+----------+--------------+-------------+----------+-----------+
| 65536 | 8192 | 65536 | 371 | 268 |
| | 8192 | 2 | 362 | 268 |
| | 8192 | 1 | 320 | 268 |
+----------+--------------+-------------+----------+-----------+
-- Thomas Zenker
c/o Lennartz electronic GmbH
Bismarckstrasse 136, D-72072 Tuebingen, Germany
Phone: +49-(0)7071-93550
Email: thz@lennartz-electronic.de
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-stable" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20011214143752.A90727>