20% cap on 100Mb XP Networks

Phandalyon

Guest
Joined
Jun 12, 2001
Messages
5,839
Has anyone heard of this?

I tried running this network tweak:

http://www.tweakxp.com/article140122.aspx

But I am still seeing a 12-16% average network useage when I look at it under the task manager while transferring large files over the local network. I have been thinking about switching to Gigabit because the network seems slow to me. But if I can actually eliminate this cap, then I may be happy with what I have. Here is the setup:

Fileserver: Windows 2000Pro, 100Mb NIC

100Mb 5 port switch

My PC: Onboard Gigabit NIC

I have tried setting it to be static 100MB full duplex on both machines and half duplex on both. I have tried the above tweak. Still I only get 12-16% utilization of my network. Is the Task manager just not reading right?

Anyone have any ideas on tweaking networks to get a little better speed out of them. I am willing to give just about any suggestion a shot.
 
Set both NIC's back to auto speed/duplex. You can't hard set these on a regular soho switch, the switch gets confused.

As long as they're both auto-negging 100/full you should get atleast 80mbit between the 2.

Those speed hacks don't work, it should work just fine out of the box.
 
I have been using auto negotiate oon both systems for about a year now and have not gotten anything better than 16% usage. I just started playing with hard settings about a week ago with no results, so they are set back to auto negotiate. I have reset everything to the defaults and have seen no results at all. I am using kind of a cheap switch, so that may be part of the issue. But I would still expect better than 16Mb transfers.
 
take the switch out of the equation and run a crossover between the two hosts...what do you get then?
 
How are you testing the performance? Doing COPY commands? dragging-and-dropping with explorer? Using a bandwith test program? Which program?

There are several knowledge-base articles which talk about network performance issues, like Q842664. Have you researched those to see if they apply?
 
Just hooked up the crossover cable and the speeds did not change at all. I have tried both directions too. I did get one spike of 20% usage, but it averaged around maybe 15%.

I am back on the switch now to get my internet connection back. It might help to know how my network is layed out, its a bit funky.

.................................DSL Modem
........................................|
....................................../...\
..................................../.......\
....................Wireless AP......5 Port Switch
...........................|..........................|
......................Pocket PC............../...\
................................................../.......\
...................................File Server........My Computer

I hope that makes sense. My File Server is also my (pseudo) HTPC. I gotta have a place to watch my south parks.

My My Docs directory on my Computer is actually a folder on the File Server, so I do a lot of transfers back and forth. Some of the tweaks I have run across have really helped load my Network directories faster. This is the most helpful tweak I have run across:

http://ko.offroadpakistan.com/computing/2003_06/winxp_network_tweak.html

But I am stuck on this speed thing. Now that we have ruled out the switch or anything else on the network, it is between these two computers. Do you think Computer speed could have an effect? My fileserver is not a great machine (Athlon 1GHz with 256MB RAM)
 
Do you have any 10Mbps devices plugged into the switch, like a router/modem?

Some cheap/crappy switches can't handle having different speeds on different ports simultaneously and cap all of them at the lowest.

I believe the term "wire speed" describes when a switch can give each port the best speed possible regardless of what's on other ports.

*edit* NM
 
mikeblas said:
How are you testing the performance? Doing COPY commands? dragging-and-dropping with explorer? Using a bandwith test program? Which program?

There are several knowledge-base articles which talk about network performance issues, like Q842664. Have you researched those to see if they apply?


I am doing drag and drop and monitoring the speed through the task manager's Network tab. Programs like Sisoft Sandra and the like all shop that I am pulling slightly below average bandwidth. My Sandra test showed 8-9MB/Sec transfer rates. That is about what I would expect, maybe even a little bit less considering I have a cheap switch.

I have searched the knowledge base and was unable to find any articles that would help. I don't have internet connection sharing or the XP firewall enabled to answer the specific example.

I do have zone alarm installed. I will try disabling that and see what happens.
 
Sorry, dumb question, but it is the 100Mbps fileserver that you are doing the monitoring from isn't it?
 
O[H]-Zone said:
What are you transferring? One big file, or a bunch of small ones?


I have tried both. I generally test it with one big file. But I get the same results with several small ones.


I did just shut down all running programs on both computers and tried a transfer. I got a spike of 50% initially and then transfers dropped to <20% when I sent from my computer to the fileserver. I got a spike of about 35% then dropped to 20% going the other way. Let me reboot and I will try the small files again.
 
MartinX said:
Sorry, dumb question, but it is the 100Mbps fileserver that you are doing the monitoring from isn't it?


Unfortunately, no. I can try it the other way around though. I don't know if Win2K Pro has the ability to monitor file transfers.

I am basically only looking at the amount of bandwith that is being used by my NIC in my XP machine.

A thought just occured to me...

I have an onboard Gigabit NIC in this machine. The connection speed is reporting at 100Mbps, which it should. Could the percentage be the % of the Gigabit? That Would not make a lot of sense, but it is a thought. It looks like it is a % of the current connection speed. Let me post a pic of my last test.
 
Phandalyon said:
I am doing drag and drop and monitoring the speed through the task manager's Network tab.

I'd suggest trying with a better copy program, like ROBOCOPY; or a network benchmark program, like NTTTCP or NETSPD.

If the problem occurrs with the low-level benchmark programs, then you know the issue is your hardware or the driver configuration. When copying files, there's lots of overhead and variance: security, the redirector, and so on.
 
mikeblas said:
I'd suggest trying with a better copy program, like ROBOCOPY; or a network benchmark program, like NTTTCP or NETSPD.

If the problem occurrs with the low-level benchmark programs, then you know the issue is your hardware or the driver configuration. When copying files, there's lots of overhead and variance: security, the redirector, and so on.

Let me take a look at those and see what I come up with.

And to answer an earlier question. My transfer rates with large groups of small files are even worse, about 12-14% utilization of the NIC.

Does anyone else have windows XP (Pro) that could dry a file transfer while monitoring their network utilization graph and see what kind of utilization they are getting?
 
Also, are you a member of a workgroup or a domain? Do you run IPSEC?
 
mikeblas said:
Also, are you a member of a workgroup or a domain? Do you run IPSEC?

Workgroup only. It is just the two computers, so I have never bothered to set up a domain or anything. I figured peer to peer would be sufficient.

Not sure about ipsec. In fact, at the moment I am wiped out enough to not even know how to check that.
 
OK, so the plot thickens. The TTCP program I downloaded:

http://www.pcausa.com/Utilities/pcattcp.htm

Shows a test of 9.2MBps. As a test I opened up my Network monitor in the Tsk manager and tested it again, it shot the graph up to 98% useage of the NIC.

So I can do a 98% useage with my hardware, at least a spike of it, which is twice as high as any spike I have seen so far. But file transfers are still at terrible speeds.

Any thoughts?
 
Oh, and the promised picture:

clipboard01.jpg


This left two peaks are a transfer and a recieve as I normally see them. The right two were from when I killed all running programs and tried. The spikes at the beginning of the transfer were the highest I had seen until I ran the ttcp program, which gave my one spike of 98%
 
how fast are the hard drives either end ?

you seriously need to get some patience lol
 
p3n said:
how fast are the hard drives either end ?

you seriously need to get some patience lol


One end is a WD Raptor (10K RPM SATA)

The other end is a 7200 RPM ATA100

Its not that I have no patience, its that I just want things to be working right. Things not working right really bugs me.
 
Crashsector said:
Do you have Windows QoS un-bound to your NIC? In fact, remove it all together.


I will give that a shot now.

edit:
No luck.

Speeds are still the same.

I am beginning to wonder if it is not just the speed of my file server. It is not a great PC to start with, it was put together with spare parts I had lying around. It is an Athlon 1000 (266 bus) with 256MB Corsair CAS2 SDRAM. It may just not be fast enough.

Has anyone else had a chance to give this a shot and see if XP gives them the same results?
 
I'm not really sure it's easy to tune file copies. The OS doesn't use multiple threads when copying to or from the network; I don't even think there's overlapped I/O. I thik you're better off tuning the cards at a low level, then assuming the OS will do its best when copying files around.

The spikes don't matter; you want consistent throughput.

If you've eliminated the switch with your crossover cable, then I think you need to move on to replacing network cards. After you've checked that you're not using IPSEC or QOS, then there aren't many other things which you might end up finding wrong in the settings.

OTOH, maybe you should tell us what all your net card settings are. If you've been jiggling them, maybe you've bumped the wrong one.

I can't give you anything to compare with because I have gigabit... but I end up with close to the theoritical maximum if I tune the parameters. (Though that seems like a lot of retransmits ... I don't remember seeing that before.)

Code:
C:\>netspd -d50k -c -t2 -o4 192.168.111.184
Connected 192.168.111.184 on port 5001
Connected 192.168.111.184 on port 5002
Starting net I/O thread #0
Starting net I/O thread #1
Ending   net I/O thread #1.
Thread[1]/interface[0] processed 00000000C8030000 bytes
Ending   net I/O thread #0.
Thread[0]/interface[0] processed 00000000C7C80000 bytes

CPU   0 usage = 99.8% (krnl=50.4% , user=49.4%)
CPU   1 usage = 99.8% (krnl=47.1% , user=52.7%)
-----------------------------------------------
CPU avg usage = 99.8% (krnl=48.8% , user=51.0%)

Total packets sent         : 482698
Total packets received     : 472402
Total re-transmits         : 4570785

Thread     MEGs      time     MB/sec     mb/sec
===============================================
   0       6706    101.04      63.29     531.01
   1       6706    101.04      63.29     531.01
-----------------------------------------------
Total     13412    101.04     126.60    1062.03
 
When I use the TTCP program I ran yesterday it shows clost to my theoretical limit as well. The XP Network Monitor reflects that, too. It shows a spike of 98% useage. The problem is that the TTCP program and Sandra and all of the other programs I have tried don't give me the option of a big enough file to actually test steady transfer rates. All they do is spike me to my max potential. Then anytime I try to transfer a file it goes slow. If that is just the way it is, then thats fine. I just want to make sure there is nothing wrong on my system. If you have XP, open up the task manager to the Networking tab and then run netspd. You will most likely see a single spike to 98 ish percent of the bandwidth useage. Then if you transfer a file see what transfer rates you are getting in the task manager and it may just answer the question about whether or not this is even a problem or if it is normal.

Here are my settings for both computers:

File Server:

Linksys LNE100TX NIC

Installed components:
Client for microsoft Networks
File and Printer Sharing
Internet Protocol (TCP/IP)

Card Configuration-Advanced:
APM Mode Wakeup - Disabled
Flow Control - Enabled
Link Wakeup (Off->On) - Enabled
Link Wakeup (On->Off) - Disabled
Magic Packet Wakeup - Enabled
Media Type - AutoSense (This is the only one I have changed)
Priority Packet - Disabled
VLAN ID - Not Present
Wake On LAN - Enabled


Workstation:

Intel 82540EM Based NIC (Onboard)

Installed Components:
Client for microsoft Networks
File and Printer Sharing
Internet Protocol (TCP/IP)

Card Configuration-Advanced:
Adaptive Inter-Frame spacing - Enabled
Coalesce Buffers - 128
Enable PME - No Action
Flow Control - Both On
Link Speed and Duplex - Auto Detect (Only one I have changed)
Locally Administered Address - Not Present
Offload Recieve TCP Checksum - On
Offload TCP Segmentation - On
Offload Transmit IP Checksum - On
Offload Transmit TCP Checksum - On
Recieve Descriptors - 256
Transmit Descriptors - 256
Wake On Settings - OS Controlled
 
Phandalyon said:
You will most likely see a single spike to 98 ish percent of the bandwidth useage.

Nope -- I rise to about 55%, then hang there until the test is done.

Phandalyon said:
Here are my settings for both computers:
Nothing jumps out at me there, though I'd probably turn off WOL if I wasn't using it. The Linksys cards don't offer much in the way of tunable parameters; as you can see, the Intel card gives you lots better control.
 
mikeblas said:
Nope -- I rise to about 55%, then hang there until the test is done.


Nothing jumps out at me there, though I'd probably turn off WOL if I wasn't using it. The Linksys cards don't offer much in the way of tunable parameters; as you can see, the Intel card gives you lots better control.


Interesting. So it is something in my systems somewhere. Or perhaps the computer speed of the file server. My PC is plenty fast. I will turn off WOL and see what happens.

edit: dasabled on both computers and no luck. I wonder if it could be the cabling between the two computers. I have wall jacks in my office and in my Living Room (Where the Fileserver is) for the network and jumpers on either end. To test the crossover, I just replaced a jumper with a crossover.

But then if it was a cable issue, that would show up in more than just speed I would imagine. Everithing is crimped as 586B connectors and I ran my cable tester from one end to the other without any errors at all.
 
OK, I just did some manual testing on the connection speed.

This is very rough at best.

Using a 10MB file and a stopwatch, I calculated rough speeds in MB and Mb.

File Server -> Workstation: 1.98MB 11.83Mb

Workstation -> File Server: 1.62MB 12.94Mb

Now as a benchmark, I also did local drives to see.

On the File Server between the two local drives (Both ATA100 7200RPM) 4.4MB 35.24Mb

On the Workstation between 2 local drives (WD Raptor and ATA133 7200RPM) instant (Too fast to measure)

The File server shows less than stellar performance, so I am starting to think that it might be the issue. I need to get another fast PC on my network and try that to see what happens I think.
 
Well, I think I can call it problem solved, or at least problem known.

I decided to run some bigger file transfer tests and came up with the following results:

252MB File:

File Server -> Workstation: 1.6 MBps 12.84 Mbps

Workstation -> File Server: 1.68 MBps 13.47 Mbps

Workstation -> Workstation: 40.32 MBps 322.56 Mbps

File Server -> File Server: .97 MBps 7.76 Mbps

65.9MB File

File Server -> Workstation: 1.6 MBps 12.8 Mbps

Workstation -> File Server: 1.62 MBps 12.95 Mbps

Workstation -> Workstation: 28.16 MBps 225.3 Mbps

File Server -> File Server: .99 MBps 7.88 Mbps


It looks like my poor old 6 year old file server is just beyond help. The transfer speeds listed are from a 40GB ATA100 7200RPM drive to a 120GB ATA100 7200RPM drive. I think I need more RAM, but I will probably just replace the machine eventually, so its not worth the $$. I may have some SDRAM lying around that I could try though.
 
That's interesting. What makes you think you need more RAM? What're the specs on the file server?

What do drive benchmarks tell you about the server? The curious thing is that NETSPD and NNTTCP don't use the disk, so they shouldn't be affected by your bad disk I/O performance.
 
mikeblas said:
That's interesting. What makes you think you need more RAM? What're the specs on the file server?

What do drive benchmarks tell you about the server? The curious thing is that NETSPD and NNTTCP don't use the disk, so they shouldn't be affected by your bad disk I/O performance.

It was just a guess on the RAM. I stuck an extra 128MB stick in that PC and it did not change anything. Now I am poking around with drive speeds. For some reason both of the HDs on that computer are racognized as SCSI when they are really IDE.

As far as the TTCP. I could not get netspd to work on my 2k machine, it was missing a dll file. The TTCP program I downloaded did show a 98% useage spike, but it did not have the ability to send enough data to produce more than a quick spike. But it did hit a high number every time I ran it. I am going to play around with that a bit more since I have time to understand it better now, but first I am going to see if I can fix this drive issue and see if that makes any difference in my network transfer speeds.

edit: oh, yeah. Drive specs.

Abit KT7A mobo
AMD Athlon 1000 (266 fsb)
364MB PC133 RAM
Maxtor ATA100 7200RPM 40GB (2MB Buffer)
WD SE ATA100 7200RPM 120GB (8MB buffer)
Liteon DVD-ROM
ATI Radeon 7000 64MB
Linksys LNE100TX
Sound Blaster Live! 5.1
 
Turns out the driver issues were the cause of the whole problem. I am now getting a 75-80% bandwidth useage out of my network screen in the task manager.

Its all fixed and I am extremely happy. Its is about 3-4 times faster now.

Not a network problem after all. I still can't figure out why those drivers weren't right though.

Oh, well. Thanks for the help!
 
Phandalyon said:
Not a network problem after all. I still can't figure out why those drivers weren't right though.

Which drivers weren't right?

Phandalyon said:
but it did not have the ability to send enough data to produce more than a quick spike.

That's confusing. Why can't you use the -b option to specify a larger buffer count and get a bigger transfer duration?

Anyway, I'm glad you've got to the bottom of it.
 
mikeblas said:
Which drivers weren't right?



That's confusing. Why can't you use the -b option to specify a larger buffer count and get a bigger transfer duration?

Anyway, I'm glad you've got to the bottom of it.


It was the IDE controller drivers that were apparently casing the problem.

I need to sit down and play with the ttcp software more to figure out all the switches and tricks, but I am going to bet that the TTCP software would have shown that my connection was just fine. So that one was most likely user error


An install of the january 2005 Via 4 in 1s was what solved the problem in the the end. I also installed the Via IDE controller drivers specifically found on Abit's site.
 
I am glad you solved it, I was going to post a new thread regarding the network utilization percentage, but instead I tried searching first (page 1 "Starting A New Thread for Dummies”) heh.

I'll try the IDE drivers first, since my server never been updated ( hardware Driver wise ) ever since it created.

Then I'll go step by step... Unless there are new program or utility that might identify my problem exactly.

Thank you again for bringing it up already and thanks to all who spread their thoughts!

( W2k3 without SP1 till 3d/9m/05 -don't ask why)
 
Back
Top