Windows 7 and Windows 8 Gigabit card throughput difference on same PC.

Forums › LAN (ethernet/Wifi/routers/Bluetooth) › Windows 7 and Windows 8 Gigabit card throughput difference on same PC.

DrStrangelove

368 posts

Ultimate Geek

#103305 3-Jun-2012 19:10

I have an old PC with Windows XP, Windows 7 and Windows 8 running on it (different partitions)

In the PC I have a TP-Link(TG-3269) PCI Gigabit Ethernet(GBE) card running a RealTek driver on a RTL8169 chip.

On the Windows 7 OS I can get a Windows cmd(dos) ftp download at ~269Mbps from my NAS.

    ftp> verboseVerbose mode On .
    ftp> get Windows8-ConsumerPreview-32bit-English.iso
    200 PORT command successful
    150 Opening BINARY mode data connection for Windows8-ConsumerPreview-32bit-English.iso (2711396352 bytes)
    226 Transfer complete
    ftp: 2711396352 bytes received in 80.52Seconds 33673.16Kbytes/sec.
    ftp>

On the Windows 8 OS I can get a Windows cmd(dos) ftp download at ~495Mbps from my NAS.

    ftp> get Windows8-ConsumerPreview-32bit-English.iso
    200 PORT command successful
    150 Opening BINARY mode data connection for Windows8-ConsumerPreview-32bit-English.iso (2711396352 bytes)
    226 Transfer complete
    ftp: 2711396352 bytes received in 43.86Seconds 61823.57Kbytes/sec.
    ftp>

Now this is right up there with 'spoon bending'... I'm amazed.
Windows 8 bandwidth is almost double that of Windows 7!!!!!!!
Have I done something REALLY stupid on my Windows 7 component configuration of the PC?

I only have the one gigabit card to test, otherwise I'd check if it's the same on other network cards/Windows 7/8 combinations.

The Hardware is all the same, even right down to 'sharing' the same disk (different partitions), so this can only be due to software and the configuration settings/drivers on the PC and the two OS (Win7, Win8).

The GBE card is running 1GB full-duplex on both systems and with no difference in the two configurations.

The Win7 driver on the GBE card is 7.58.411.2012 (4/11/2012)
The Win8 driver on the GBE card is 8.1.1019.2011 (19/10/2011)

In windows 8 the Ethernet connection has to be up and running before you even attempt to login and is seen as active, ready and waiting. This allows network login accounts.

I have checked Windows 7 disk write-cache and it's enabled.

So, what the fluffy duck did Microsoft do to Windows 8 to almost double the network speed of a GBE card???

Any thoughts? Is there a disk write performance tweak, I've missed. It's not something silly like windowing or buffering in the ftp setup within cmd window?

gzt

17160 posts

Uber Geek

Lifetime subscriber

#635177 3-Jun-2012 19:59

Looks like Win 8 is using an earlier driver version. Maybe the later version you are running on 7 is something that WU picked for you but is not correct. Try the earlier version on your 7 box.

DrStrangelove

368 posts

Ultimate Geek

#635188 3-Jun-2012 20:18

Looked at that ealier myself. And it's a good point. I Had an early version (blessed by Microsoft driver update) originally and then tried a few later ones right up to the latest release from Realtek. All give the same slower throughput on a Win7 system.

Interestedly I believe the 8.xxx denote Windows 8 and the 7.xxxxxx denotes Windows 7.
Vista appears to be 6.xxxxx and Win XP is 5.xxxxxxx

So the 8.1.xxxx driver is more likely to be a driver bespoke to Windows 8

I did try looking for the Windows 8 driver and then attempting to ram it down the throat of Win7, BUT I couldn't find the driver.
Not even on Realtek software download.

It's a mystery. Could attempt to 'pull' the .dll and .inf off Win8 and 'fudge' it on Win7, but thought I'd ask a few questions before I got the big hammer out.

Had a look at regedit too for the GBE interface controlSet class setting and there wasn't anything that popped out and looked mega different between the two systems, drivers aside.

PC software is not my strong point so I'm erring on the side of Oh I didn't set PC setting x for disk setting x.

I just can't see how one could get an old PCI network card and almost double the speed on the same media...

gzt

17160 posts

Uber Geek

Lifetime subscriber

#635198 3-Jun-2012 20:28

For a sanity check before the big guns, maybe check the numbers on a different transfer protocol and/or something from a 3rd party vendor.

DrStrangelove

368 posts

Ultimate Geek

#635263 3-Jun-2012 23:04

Cool. Any thoughts on what would be a good test. I use ftp as it gives the best stats.

Anything else that would give reliable results.

Maybe I could run a transfer over PPTP as the server is the first hop router.

Downloading tons of LAN stress tools and doing some disk error checking.

Hopefully as you suggest these stress test will give a broader view of what's happening.

insane

3242 posts

Uber Geek

ID Verified

Trusted

#635284 4-Jun-2012 00:48

use IPERF or JPERF if you want a GUI, Will need two PC's though

Run a UDP test I suspect the numbers will be similar.

If TCP results are where you see the difference then perhaps they have further enhanced the TCP window scaling in Win8....was actually already pretty good in Win7 / server 2K8.

DrStrangelove

368 posts

Ultimate Geek

#635288 4-Jun-2012 00:56

Just did a LAN Speed Test.

This Reads and Writes data to a network share(NAS) from memory.

This cuts out any local HDD I/O on the PC.

Windows 7

Windows 8

This test shows that Windows 8 is still almost twice as fast as Windows 7 when reading from a NAS folder.
The network writes are within ball park figures of each other which does point to a read issue on the Win7 inbound.

I'm now thinking... what is parsing/vetting the inbound data on the GBE card... virus scanner/firewall maybe.. more investigation required. Maybe I have something reading the port data on Win7 that I don't on Win8.

DrStrangelove

368 posts

Ultimate Geek

#635291 4-Jun-2012 01:09

insane: use IPERF or JPERF if you want a GUI, Will need two PC's though

Run a UDP test I suspect the numbers will be similar.

If TCP results are where you see the difference then perhaps they have further enhanced the TCP window scaling in Win8....was actually already pretty good in Win7 / server 2K8.

Cool, looked at IPERF and JPERF will probably download it and have a look at it too.

UDP and TCP over a (almost point-to-point LAN with only router between NAS) point to point LAN would probably be similar.

Think I'll just do some quick tests by removing any firewall and virus checkers on windows 7 and see what happens.

Thanks for your suggestions.

Sharesies

Trade NZ and US shares and funds with Sharesies (affiliate link).

DrStrangelove

368 posts

Ultimate Geek

#635990 5-Jun-2012 12:38

Went to plan B.

Installed a clean version of Windows 7 SP1 on a new partition and used the following Microsoft installed GBE driver.
Windows firewall was enabled, but Windows Defender was disabled and no anti-virus was install for the clean build test.

TP-Link TG-3269 PCI GBE using Realtek chip

Driver Provider: Realtek
Driver Date: 10/06/2011
Driver Version: 7.46.610.211
Digital Signer: Microsoft

This gives us 497.11Mbps read throughput as seen using LAN Speed Test which is on a par with what we see on Windows 8 using a different partition on the same system/same card.

Windows 7 data read transfer rate over Gigabit Ethernet

Using a 'raw' ftp transfer we can see anywhere from 671Mbps with averages around 630Mbps on the Windows 7 partition.

This is data being read from a baby NAS (QNAP TS-112) and written to the local HDD on the desktop PC.

This is good. I'd be bold enough to say this is VERY good and may be starting to peak on the limitations of the QNAP TS-112 disk read access.

The data read bandwidth we see on the Windows 7 partition is almost the same as the Windows 8 partition, give or take the odd Mbps.

I have installed a collection of 'standard' applications on the Windows 7 'test' partition and the Gigabit bandwidth has not been impacted.

And that's where the happiness seems to end.

The Windows 8 partition is using a well known and popular Anti-virus package and returning a solid ~630Mbps read bandwidth rate. That's good.. for Windows and a QNAP TS-112 NAS.

If that same well known and popular Anti-virus package is applied to Windows 7 the data rate drops down by up to 150Mbps.

I only use the Anti-virus with a select subset of real time mail and file scanning.

If that same Anti-virus is uninstalled the bandwidth does not improve in fact it appears to drop even further. The only way I was able to recover the former bandwidth rate was to use a restore point prior to the Anti-virus install/uninstall.

So, knowing this, I am able to 'plan' an install strategy, using my 'other' Windows 7 partition as a test environment so that I'm not divesting myself of data throughput at the Ethernet interface by installing random applications and then finding it's reduced my Ethernet throughput and not having a clear backout procedure to recover said bandwidth.

Note to self - Write change control.

This is a very uneducated approach as I'm sure a smarter person would be able to identify drivers and config file which would impact on the Ethernet interface and know how to correct them if they do.

So, in summary my findings are that third party applications could impact 'adversely' on the Gigabit Ethernet interface throughput and that by just uninstalling the application there is every change you won't recover that lost bandwidth without doing a system restore to a previous point.

All my data is centralised on the NAS with Windows XP/7/8 acting as a thin client with Libraries/folders being 'linked' as junction points and seen as local folders, thus the requirement to maintain the best throughput possible in a network environment.

gzt

17160 posts

Uber Geek

Lifetime subscriber

#636314 5-Jun-2012 20:24

Nice work! Clearly the AV package is not correctly uninstalling it's nasty hooks either.

If you have time it may be worth trying other AV trial packages. This may be an 7 architectural limitation which AV packages do their best to work with or it may be an issue with just one.

mercutio

1392 posts

Uber Geek

#636324 5-Jun-2012 20:34

fwiw i get 975 megabit with windows 8 and 9k mtus with realtek onboard.

hmm actually it seems to have gone down...

PS C:\util\iperf-2.0.5-cygwin> .\iperf.exe -s
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size: 64.0 KByte (default)
------------------------------------------------------------
[ 4] local 192.168.1.42 port 5001 connected with 192.168.1.7 port 52042
[ ID] Interval Transfer Bandwidth
[ 4] 0.0-10.0 sec 1.12 GBytes 964 Mbits/sec

mercutio

1392 posts

Uber Geek

#636327 5-Jun-2012 20:38

DrStrangelove:
insane: use IPERF or JPERF if you want a GUI, Will need two PC's though

Run a UDP test I suspect the numbers will be similar.

If TCP results are where you see the difference then perhaps they have further enhanced the TCP window scaling in Win8....was actually already pretty good in Win7 / server 2K8.

Cool, looked at IPERF and JPERF will probably download it and have a look at it too.

UDP and TCP over a (almost point-to-point LAN with only router between NAS) point to point LAN would probably be similar.

Think I'll just do some quick tests by removing any firewall and virus checkers on windows 7 and see what happens.

Thanks for your suggestions.

i've had problems doing high speed udp in windows with iperf - i dunno if it's iperf or windows. linux is a lot more stable with higher udp loads.

mercutio

1392 posts

Uber Geek

#636333 5-Jun-2012 20:40

insane:

If TCP results are where you see the difference then perhaps they have further enhanced the TCP
window scaling in Win8....was actually already pretty good in Win7 / server 2K8.

i think ctcp is still off by default.

TCP Global Parameters
----------------------------------------------
Receive-Side Scaling State : enabled
Chimney Offload State : disabled
NetDMA State : disabled
Direct Cache Access (DCA) : disabled
Receive Window Auto-Tuning Level : normal
Add-On Congestion Control Provider : none
ECN Capability : disabled
RFC 1323 Timestamps : disabled
Initial RTO : 500
Receive Segment Coalescing State : disabled

yip..

you can set the initial rto lower at least though :)

"netsh int tcp set global initialrto=600" for instance.. 1000 is a reasonable setting.. 3000 is the old standard and way too high.. 500 could cause extra retransmissions for sites over 500 msec away. (most likely with bittorrent traffic to congested hosts)

hmm it won't let me enable ctcp

Regs

4066 posts

Uber Geek

Trusted

Snowflake

#636827 6-Jun-2012 17:39

I've watched a popular cheap/free av solution kill a server, even when all A/V services disabled. it only returned back to normal once this product was completely uninstalled. First thing I demand these days when performance is an issue is removal of any antivirus software.

example:
sql system A: 100 inserts per second (no antivirus, virtual machine on desktop class box)
sql system B: 2 inserts per second (antivirus A, physical machine, 10 SCSI disks RAID-5)

we ran a lot of tests and eventually saw that raw disk write performance was like 2MB/sec instead of > 100MB/sec.

sometimes, you get what you pay for. i'd be super careful about the A/V package you install, as it can be a massive chokepoint: for disk writes, network writes, URL filtering, JavaScript scanning, etc etc

Sales Engineer
Snowflake
www.snowflake.com

about.me/nzregs
Twitter: @nzregs

DrStrangelove

368 posts

Ultimate Geek

#639199 11-Jun-2012 18:37

Rebuilt my 'primary' Win7 partition and all seems to be going swimmingly on my PCI TG-3269 GBE interface

Installed iperf on my QNAP NAS and did a quick test.

Z:\Network LAN Test> iperf -s
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size: 8.00 KByte (default)
------------------------------------------------------------
[212] local 172.nn.nn.161 port 5001 connected with 172.nn.nn.10 port 43211
[ ID] Interval     Transfer    Bandwidth
[212] 0.0-10.0 sec   999 MBytes   837 Mbits/sec

Z:\Network LAN Test>

Installed Malwarebytes as a manual download virus checker and that hasn't impacted my GBE interface.

On a 'raw' ftp download from my NAS I get anything up to 657Mbps (674Mbps with TCP Global parameters changed from default)

Had a check on the Win8 partition and note that the TCP Global setting are pretty generic.

Windows 8

TCP Global Parameters
----------------------------------------------------
Receive-Side Scaling State              : enabled
Chimney Offload State                   : disabled
NetDMA State                      : disabled
Direct Cache Access (DCA)               : disabled
Receive Window Auto-Tuning Level   : normal
Add-On Congestion Control Provider : none
ECN Capability                                : disabled
RFC 1323 Timestamps                    : disabled
Initial RTO                                     : 3000
Receive Segment Coalescing State   : disabled
-----------------------------------------------------
Many of the Win8 TCP interface parameters 'can not be set on this platform'..etc, etc.

Windows 7

On my Win7 partition, I've just gone mental and even enabled experimental windowing....
Yes, experimental windowing that goes black when there danger outside, so you can't see it.

TCP Global Parameters
------------------------------------------------------
Receive-Side Scaling State            : enabled
Chimney Offload State                 : enabled
NetDMA State                      : enabled
Direct Cache Acess (DCA)                 : enabled
Receive Window Auto-Tuning Level    : experimental
Add-On Congestion Control Provider : ctcp
ECN Capability                                : enabled
RFC 1323 Timestamps                    : disabled
-------------------------------------------------------

This has given a noticeable ~20Mbps bandwidth increase.
BUT from my experience, it's like going back in time and stepping on a butterfly, where everything then changes... and mostly not for the good.

LAN Speed Test

LAN test data throughput is indicating good throughput (PCI card considered)

VirtualBox

One other thing I have noticed, VirtualBox impacts on the network throughput.
I tried installing it with none of the network adapters (bridge and Host networks), but the network interface still takes a hit.
One good thing, it only appears to decrease the Ethernet bandwidth while it is being used.

This is acceptable, given the focus is mostly on the VirtualBox client(s) at the time and when you have finished using those systems, then network normality is restored. Boo Yaa.