Performance of NDIS layer: XP vs Win7

Hi,

We are seeing a huge difference between the throughput values of a NDIS 5.1 driver on Windows XP SP3 32 bit and same driver on Windows 7 32 bit. All the parameters of the driver are exactly same, but Win7 throughput is much better.
Does this mean that Windows 7 network stack or NDIS layer has a considerable improvement over Windows XP?

We also noticed exactly same throughput figures on Windows 7 for NDIS 5.1 driver and NDIS 6.0 driver. Looks like NDIS 6.0 doesn?t give much performance gain for 100 Mbps network.

Regards,
Suresh

Out of curiosity, how are you measuring ‘throughput’? What is your test
stimulus and metric(s)?

I am perhaps incorrectly inferring from your post that you have somehow
eliminated the effects of all of the other network performance improvements
that went into NT6 and have isolated the delta-throughput improvement of
just NDIS6 providing ‘port’ services to an NDIS5 miniport.

As for the delta between an NDIS5.1 and NDIS6 miniport for the same NIC on a
100Mbps network … gosh, what do you expect? Both of those ‘drivers’
architectures on modern CPU/system hardware are capable of burying a 100Mbps
link and still have enough cycles left to play Doom, decode some YouTube
video, etc. The *link* is the performance bottleneck. It swamps any
measurement you could possibly make and thus finding the ‘improvement’ in
NDIS6 would be very very hard to get above the threshold of just what the
link is contributing to the test.

Performance testing a ‘subsystem’ like NDIS or a ‘module’ like a particular
miniport is a very challenging endeavor both in creating the test bed and
instrumentation *and* interpreting the results. It is not running iPerf or
TTCP and looking at the results as ‘the answer’.

What is your goal here? To decide if you should bother with a NDIS6
Miniport? The answer is yes.

Good Luck,
Dave Cattley

-----Original Message-----
From: xxxxx@lists.osr.com
[mailto:xxxxx@lists.osr.com] On Behalf Of
xxxxx@yahoo.com
Sent: Tuesday, February 28, 2012 2:01 AM
To: Windows System Software Devs Interest List
Subject: [ntdev] Performance of NDIS layer: XP vs Win7

Hi,

We are seeing a huge difference between the throughput values of a NDIS 5.1
driver on Windows XP SP3 32 bit and same driver on Windows 7 32 bit. All the
parameters of the driver are exactly same, but Win7 throughput is much
better.
Does this mean that Windows 7 network stack or NDIS layer has a considerable
improvement over Windows XP?

We also noticed exactly same throughput figures on Windows 7 for NDIS 5.1
driver and NDIS 6.0 driver. Looks like NDIS 6.0 doesn?t give much
performance gain for 100 Mbps network.

Regards,
Suresh


NTDEV is sponsored by OSR

For our schedule of WDF, WDM, debugging and other seminars visit:
http://www.osr.com/seminars

To unsubscribe, visit the List Server section of OSR Online at
http://www.osronline.com/page.cfm?name=ListServer

“Huge” usually has different meanings to different individuals. To me, it
sounds like your ndis51 driver and/or your XP setup have serious problems.
What are the two numbers?

Calvin

On Mon, Feb 27, 2012 at 11:01 PM, wrote:

> Hi,
>
> We are seeing a huge difference between the throughput values of a NDIS
> 5.1 driver on Windows XP SP3 32 bit and same driver on Windows 7 32 bit.
> All the parameters of the driver are exactly same, but Win7 throughput is
> much better.
> Does this mean that Windows 7 network stack or NDIS layer has a
> considerable improvement over Windows XP?
>
> We also noticed exactly same throughput figures on Windows 7 for NDIS 5.1
> driver and NDIS 6.0 driver. Looks like NDIS 6.0 doesn?t give much
> performance gain for 100 Mbps network.
>
> Regards,
> Suresh
>
>
> —
> NTDEV is sponsored by OSR
>
> For our schedule of WDF, WDM, debugging and other seminars visit:
> http://www.osr.com/seminars
>
> To unsubscribe, visit the List Server section of OSR Online at
> http://www.osronline.com/page.cfm?name=ListServer
>

Thanks Dave. I did not intend to downgrade NDIS6 or question its adaptability. That is where the future is and it is a must if you want WHQL certification. That was just a side-observation and my main goal was to find out how bad XP throughput is with respect to Win7 with the same driver.

Sorry Calvin, I should have put the numbers already. I used iperf, WSTTCP, Chariot and manual timing of file copy methods for measuring the throughput. Both Windows XP and Windows 7 are installed on the same PC and both are fresh installation.
Here are the numbers:
File copy of 300 MB = WinXP (TX: 49 sec, RX: 48 sec) - Win7 (TX: 29 sec, RX: 28 sec)
Chariot = WinXP (TX: 77 Mbps, RX: 73 Mbps) - Win7 (TX: 90 Mbps, RX: 88 Mbps)
WSTTCP = WinXP (TX: 69 Mbps, RX: 68 Mbps) - Win7 (TX: 95 Mbps, RX: 94 Mbps)

  • Suresh

Suresh,

Your numbers are awfully low if you’re testing a gigabit adapter on a
gigabit link. For WSTTCP (I’m not a big fan of it) alike test, my drivers
can easily saturate a gig or 10Gbps link without turning on TCP chimney
offload on a modern machine. i.e. 948~950 Mbps is a typical TCP
payload throughput I would expect to see with netperf on a gigabit link.

Chariot is complex. Without knowing your script, configuration, test
method, it’s hard to tell what is going on.

In my experience, performance difference between ndis5.1 and ndis6x drivers
is not practically observable on a modern computer if the there is no
advanced hardware features are enabled on the ndis6x driver. i.e. TCP
chimney offload, IPSec AES offload, RSS features which are not supported in
ndis5.1 may offer significant boost in some configurations.

Calvin
P.S. I have not paid much attention to WinXP since last couple years.

On Tue, Feb 28, 2012 at 10:17 PM, wrote:

> Thanks Dave. I did not intend to downgrade NDIS6 or question its
> adaptability. That is where the future is and it is a must if you want WHQL
> certification. That was just a side-observation and my main goal was to
> find out how bad XP throughput is with respect to Win7 with the same driver.
>
> Sorry Calvin, I should have put the numbers already. I used iperf, WSTTCP,
> Chariot and manual timing of file copy methods for measuring the
> throughput. Both Windows XP and Windows 7 are installed on the same PC and
> both are fresh installation.
> Here are the numbers:
> File copy of 300 MB = WinXP (TX: 49 sec, RX: 48 sec) - Win7 (TX: 29 sec,
> RX: 28 sec)
> Chariot = WinXP (TX: 77 Mbps, RX: 73 Mbps) - Win7 (TX: 90 Mbps, RX: 88
> Mbps)
> WSTTCP = WinXP (TX: 69 Mbps, RX: 68 Mbps) - Win7 (TX: 95 Mbps, RX: 94 Mbps)
>
> - Suresh
>
>
> —
> NTDEV is sponsored by OSR
>
> For our schedule of WDF, WDM, debugging and other seminars visit:
> http://www.osr.com/seminars
>
> To unsubscribe, visit the List Server section of OSR Online at
> http://www.osronline.com/page.cfm?name=ListServer
>

Calvin Guan (news) wrote:

In my experience, performance difference between ndis5.1 and ndis6x
drivers is not practically observable on a modern computer if the there
is no advanced hardware features are enabled on the ndis6x driver. i.e.
TCP chimney offload, IPSec AES offload, RSS features which are not
supported in ndis5.1 may offer significant boost in some configurations.
Calvin
P.S. I have not paid much attention to WinXP since last couple years.

Other thing to consider is that in WinXP the TCP stack is not configured
to use large windows sizes unless the user tweaks the registry. This can
lead to low performance on faster link speeds as well.

Mike

Mike Pumford, Senior Software Engineer
MPC Data Limited
e-mail: xxxxx@mpcdata.com web: www.mpcdata.com
tel: +44 (0) 1225 710600 fax: +44 (0) 1225 710601
ddi: +44 (0) 1225 710635