[linux-elitists] USB-Serial and pppd

Greg Folkert greg@gregfolkert.net
Thu Jun 12 20:04:19 PDT 2008


On Thu, 2008-06-12 at 16:32 -0700, Greg KH wrote: 
> On Thu, Jun 12, 2008 at 03:45:25PM -0400, Greg Folkert wrote:
> > 
> >         linux-image-2.6.25-2-686-bigmem v2.6.25-4
> >         linux-image-2.6.24-1-686-bigmem v2.6.24-6
> 
> That's good, but you should update your .24 one to .25 to make sure.
> 
> > I use the 2.6.24 kernel as ACPI/APM/CPUFREQ is screwed up on the 2.6.25
> > kernel, I only get 800MHz out of this laptop with the 2.6.25 kernel
> > running along with the cpu frequency scaling stuff.
> 
> You might want to just turn that off then :)

Yeah, but when you are using a laptop... the fans start to turn on and
get really noisy running at full speed all the time. Plus Battery life
tends to be more on the suckish side of things.

> > I had an extended stint in a colo in Denver about 300ft away from a
> > Verizon Cell tower and only got 512K/sec in Linux period, the guy next
> > to me had a MacBookPro with a ExpressCard model and same 512K results
> > came along in my Lenovo using that card. He gets hyarge throughput on
> > his MacBookPro compared to my T61. I tried on 2.6.24 only as I hadn't
> > had the time to update to 2.6.25 when it came out. But now it gives
> > similar results.
> 
> Ok, I can't understand what you are saying here, is it that you got
> 512K/second just like someone else using os-x did?  If so, what is the
> problem?

I meant to say that he lent me his card and I got the exact same results
in Linux 512K/sec with my AC595 or his AC597. And Similar 2M/sec in
WindowsXP with either one.


2M/sec for OSX at the using the ExpressCard device in His machine,
512K/sec for that same card in my machine using Linux, 2M/sec for the
same card in my machine using WindowsXP.

> Also note, I forgot to mention this the first time, but some people's
> data plans cap the throughput to a specific level.  I proved this to
> someone by using my card in their laptop with the exact same driver and
> device type.  My data plan has no caps, his did.
> 
> So, you might want to check with your phone company :)

My AC595 card has no cap in WindowsXP, neither does the AC597
ExpressCard device from my friend.

My AC595 has been as high as 2M/sec with WindowsXP.

Never higher than 512K/sec with Linux (with my Lenovo T61 stinkpad or my
Dell Inspion 8000)

> > And its frustrating, but I'm in this for the long haul no matter what
> > SCOg thinks, or what Law.com mis-believes about the GPL and IP. Or wrong
> > thinking/attacks by Microsoft.
> 
> And this is related to throughput of a driver how?  Please...

I am just saying people give up and never FINISH a problem like this.
I've seen lots of starts but no finishing, either because "Oh its fixed,
no more need to biatch about it" or they leave and never return.

> Ok, please measure up and down throughput separately, as they are very
> different.  If you can rebuild your kernel, here's some things you can
> try to change these numbers.

I've only really speed tested in Linux, everything else was transfers to
my work and back... telling my the how long it takes to D/L or U/L these
files and data.

Using Speedtest.net, from my REALLY CRAPPY signal area tonight:
In Linux I get 490-512K/sec Download, 290-310K/sec Upload, nearly every
time.

In WindowsXP from the same location with CRAPPY signal I get this, on
the same machine:

I get 980-1080K/sec Download, 280-320K/sec Upload

> In the file, drivers/usb/serial/sierra.c, make sure that the buffer
> sizes and numbers all show:
> 	#define N_IN_URB        4
> 	#define N_OUT_URB       4
> 	#define IN_BUFLEN       4096

Debian's kernel is set there. A recompile later  with URB at 8 (and
turning off the CPUFREQ stuff for my laptop in 2.6.25) and we are stuck
at at the same level as with 2.6.24.

> If not, increase them to those levels.
> 
> If they are set to those levels, then you can try to increase the number
> of URBs from 4 to 8, but it's usually a level of diminishing returns,
> the network is usually too slow that you really need more than that big
> of a buffer in flight at once.
> 
> But try it and let me know what happens.  And use 2.6.25, no one cares
> about 2.6.24 anymore :)

Yeah yeah, I know.

I've got 2.6.25 straight Debian with URB of 4, getting 512K max D/L

I've got 2.6.25 with a modified sierra.c with a URB of 8, getting 512K
max D/L

Of course Straight Debian 2.6.24 with URB at 4,, getting 512K max D/L

I've even put my External Antenna on the Card, with a 6db gain... same
results.

Tonight with my crappy connection, In Windows I Downloaded Puppy Linux
v4.0 (Dingo) and averaged 1010K/sec but as high as 1670k/sec and as low
as 90K/sec.

In Linux (on 2.6.25 modified), I downloaded the exact same file,
averaged 501K/sec with a maximum speed of 512K and a minimum speed of
70K.

Something doesn't seem right. What do you need from me?
-- 
greg@gregfolkert.net
PGP key 1024D/B524687C 2003-08-05
Fingerprint: E1D3 E3D7 5850 957E FED0  2B3A ED66 6971 B524 687C
Alternate Fingerprint: 09F9 1102 9D74  E35B D841 56C5 6356 88C0
Alternate Fingerprint: 455F E104 22CA  29C4 933F 9505 2B79 2AB2
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
Url : http://allium.zgp.org/pipermail/linux-elitists/attachments/20080612/bda432ec/attachment.pgp 


More information about the linux-elitists mailing list