Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 26 Jan 2007 15:36:48 +0100
From:      Hans Petter Selasky <hselasky@c2i.net>
To:        freebsd-current@freebsd.org
Cc:        Alexey Karagodov <karagodov@gmail.com>, Stefan Ehmann <shoesoft@gmx.net>
Subject:   Re: Interesting speed benchmarks
Message-ID:  <200701261536.48893.hselasky@c2i.net>
In-Reply-To: <200701261341.03742.shoesoft@gmx.net>
References:  <20070125.192448.-432840241.imp@bsdimp.com> <cb5206420701260435s66e0687bnb467a42379d0a8d3@mail.gmail.com> <200701261341.03742.shoesoft@gmx.net>

next in thread | previous in thread | raw e-mail | index | archive | help
On Friday 26 January 2007 13:41, Stefan Ehmann wrote:
> On Friday 26 January 2007 13:35, Andrew Pantyukhin wrote:
> > On 1/26/07, Stefan Ehmann <shoesoft@gmx.net> wrote:
> > > On Friday 26 January 2007 11:00, Alexey Karagodov wrote:
> > > > what manufacturer says about usb speeds?
> > > > that is the question
> > >
> > > Well, "up to 56MB/s" which is pretty much full USB2 speed.

It is called high speed USB, and it can go up 53 MB/s with a payload of 512 
bytes per packet according to "Table 5-10. High-speed Bulk Transaction 
Limits" in the USB 2.0 specification. The table does not say anything about 
whether this include bit-stuffing or not. If bit stuffing is not included, 
then you have to divide this value by 1.20 approximately for the worst case, 
all 1's. 53 MB/s div 1.20 = 44 MB/s.

> > >
> > > But writing it on the box doesn't mean the speed can actually be
> > > reached.
> > >
> > > Benchmarking on windows might be interesting, but I don't know how to
> > > measure raw disk io on windows.
> >
> > Format the disk, copy a large file to/from it, divide
> > its size by time spent, add the word "approximately" :-)
>
> I'd rather not format a drive with my backups and other stuff on it :-)

Results with the new USB stack*:

Changing the interrupt delay from 2 microframes to 1 microframe gave me 
2MBytes more per second on the EHCI controller.

I connected two high speed "umass" capable devices to the same EHCI controller 
on my computer, and did a "dd" on both devices at the same time, with a block 
size of 131072 bytes.

The one device transferred 22 MB/s. The other device transferred 16 MB/s. 
Summed up this yields 38 MB/s. Used alone these devices can transfer 27 MB/s 
and 20 MB/s. It seems clear that the EHCI controller is saturated at 38 MB/s. 

%dmesg |grep ehci
ehci0: <Intel 82801DB/L/M (ICH4) USB 2.0 controller> mem 0xe0100000-0xe01003ff 
irq 10 at device 29.7 on pci0
usb3: <Intel 82801DB/L/M (ICH4) USB 2.0 controller> on ehci0
%

--HPS

* http://www.turbocat.net/~hselasky/usb4bsd or see the FreeBSD-perforce USB 
project.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200701261536.48893.hselasky>