Prev: Prioritized RAW Ethernet
Next: how to notify the user of a certain event from within a NDIS IM?
From: suresh on 5 Feb 2010 02:23
Hi,

We have a usb based Ethernet adapter. It is a high speed (2.0) device.
The throughput observed on Win7 is around 30 Mbps lower than that
observed on Windows XP!
Both drivers are certified. There is not much difference between these
drivers except NDIS5-NDIS6 and WDM-WDF related changes. Even the same
NDIS5-WDM driver used on XP gives lower throughput figures on Win7.

When checking the receive side operations we noticed that the
difference between 2 consecutive Read completions on XP is around
30-50 microseconds, while the same on Win7 is around 200 microseconds.
We are using Continuous Reader with default (NumPendingReads = 0)
value. The observations didn't change when using different (upto 10)
values, or using manual polling based reading.

Since the traffic generator on the other side is same in both cases,
the Win7 is also expected to receive those packets as fast as XP.
Is this an expected behavior or I am missing something here?
Thanks,

Regards,
Suresh

From: Doron Holan [MSFT] on 5 Feb 2010 13:57
you can use the new ETW tracing facilities for USB to see what is happening
in win7

d

--

This posting is provided "AS IS" with no warranties, and confers no rights.


"suresh" <patil_suresh420(a)yahoo.com> wrote in message
news:2271bd0d-a08f-4525-946d-f7e35774159a(a)a5g2000prg.googlegroups.com...
> Hi,
>
> We have a usb based Ethernet adapter. It is a high speed (2.0) device.
> The throughput observed on Win7 is around 30 Mbps lower than that
> observed on Windows XP!
> Both drivers are certified. There is not much difference between these
> drivers except NDIS5-NDIS6 and WDM-WDF related changes. Even the same
> NDIS5-WDM driver used on XP gives lower throughput figures on Win7.
>
> When checking the receive side operations we noticed that the
> difference between 2 consecutive Read completions on XP is around
> 30-50 microseconds, while the same on Win7 is around 200 microseconds.
> We are using Continuous Reader with default (NumPendingReads = 0)
> value. The observations didn't change when using different (upto 10)
> values, or using manual polling based reading.
>
> Since the traffic generator on the other side is same in both cases,
> the Win7 is also expected to receive those packets as fast as XP.
> Is this an expected behavior or I am missing something here?
> Thanks,
>
> Regards,
> Suresh
>
From: Philip Ries [MSFT] on 5 Feb 2010 15:31
Get started here:

http://blogs.msdn.com/usbcoreblog/archive/2009/12/04/etw-in-the-windows-7-usb-core-stack.aspx
http://blogs.msdn.com/usbcoreblog/archive/2009/12/21/answering-the-question-what-s-wrong-with-my-device-using-usb-etw.aspx

Doron Holan [MSFT] wrote:
> you can use the new ETW tracing facilities for USB to see what is
> happening in win7
>
> d
>
From: suresh on 12 Feb 2010 02:49
Thanks Doron and Philip,

This tools is really useful and gives good amount of insight into USB
operations. I am still going through the logs to analyze them further.

In the meantime I had taken a capture of usb bus analyzer on XP and on
Win7 (using continuous reader).
On XP, a bulk-in is always seen ready and we see NAKs when there is no
incoming data.
On Win7 however, sometimes there is no bulk-in pending for more than
2-3 microframes! I tried different values for NumPendingReads
parameter from 0 to 32, but the gap is always there. Even the manual
polling method with as much as 32 reads pending with usb gives this
gap (approximately once in 10 microframes).

XP usb performance is clearly better than Win7, at least the read
related one. Is this a known fact? Is there any parameters like
NetworkThrottlingIndex I found earlier with respect to network
performance in Win7 (Ref. http://support.microsoft.com/kb/948066)?

Regards,
Suresh

On Feb 6, 1:31 am, "Philip Ries [MSFT]" <phr...(a)microsoft.com> wrote:
> Get started here:
>
> http://blogs.msdn.com/usbcoreblog/archive/2009/12/04/etw-in-the-windo...http://blogs.msdn.com/usbcoreblog/archive/2009/12/21/answering-the-qu...
>
>
>
> Doron Holan [MSFT] wrote:
> > you can use the new ETW tracing facilities for USB to see what is
> > happening in win7
>
> > d- Hide quoted text -
>
> - Show quoted text -

 | 
Pages: 1
Prev: Prioritized RAW Ethernet
Next: how to notify the user of a certain event from within a NDIS IM?