Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.os.linux.networking > #630 > unrolled thread

VERY SLOW Intel e1000 1GB NIC

Back to article view | Back to comp.os.linux.networking

Contents

  VERY SLOW Intel e1000 1GB NIC Kevin the Drummer <nobody@cosgroves.us> - 2011-09-23 00:50 -0500
    Re: VERY SLOW Intel e1000 1GB NIC Jorgen Grahn <grahn+nntp@snipabacken.se> - 2011-09-23 09:14 +0000
      Re: VERY SLOW Intel e1000 1GB NIC Kevin the Drummer <nobody@cosgroves.us> - 2011-09-23 18:15 -0500
        Re: VERY SLOW Intel e1000 1GB NIC Jorgen Grahn <grahn+nntp@snipabacken.se> - 2011-09-24 17:33 +0000
          Re: VERY SLOW Intel e1000 1GB NIC Kevin the Drummer <nobody@cosgroves.us> - 2011-09-24 19:30 -0500
            Re: VERY SLOW Intel e1000 1GB NIC david <none@nospam.com> - 2011-09-25 11:48 +0000
              Re: VERY SLOW Intel e1000 1GB NIC ein <ein@no.spam> - 2011-09-27 23:33 +0200
                Re: VERY SLOW Intel e1000 1GB NIC Jorgen Grahn <grahn+nntp@snipabacken.se> - 2011-09-27 21:58 +0000
                  Re: VERY SLOW Intel e1000 1GB NIC ein <ein@no.spam> - 2011-09-28 10:50 +0200
    Re: VERY SLOW Intel e1000 1GB NIC David Schwartz <davids@webmaster.com> - 2011-09-24 14:28 -0700
      Re: VERY SLOW Intel e1000 1GB NIC Kevin the Drummer <nobody@cosgroves.us> - 2011-09-24 19:34 -0500
        Re: VERY SLOW Intel e1000 1GB NIC Kevin the Drummer <nobody@cosgroves.us> - 2011-09-27 23:40 -0500
          Re: VERY SLOW Intel e1000 1GB NIC ein <ein@no.spam> - 2011-09-28 10:56 +0200
            Re: VERY SLOW Intel e1000 1GB NIC Kevin the Drummer <nobody@cosgroves.us> - 2011-09-29 23:56 -0500
    Re: VERY SLOW Intel e1000 1GB NIC Kevin the Drummer <nobody@cosgroves.us> - 2011-09-24 19:40 -0500

#630 — VERY SLOW Intel e1000 1GB NIC

I bought an Intel 1GB network card to replace my 100MB Netgear
card.  My network speed has DRAMATICALLY decreased.  My network
monitor shows bursts of ~100KB traffic and then pauses.  This
pattern repeats constantly, and between machines in the house,
not over my DSL.  I searched for info on this issue and found
several people reported similar experiences over the years.  Many
also experienced high CPU utilization.  But, my CPU is near 2%
used.

My house is 1GB over CAT6 cables.  All of the networking hardware
outside my computer cases is Netgear stuff, switches & routers.
The Intel card uses the e1000 kernel module in a 2.6.17.14
kernel.  The specific card I'm using is an "Intel PWLA8391GT
Desktop Adapter PRO/1000 GT 10/100/1000Mbps PCI"

What might I do to tune up this card for full GB performance?
Should I just return it and get another card?  I bought Intel
thinking that would work, and because it was highly rated on
the shopping site I visited.

Thanks...

-- 
PLEASE post a SUMMARY of the answer(s) to your question(s)!
Unless otherwise noted, the statements herein reflect my personal
opinions and not those of any organization with which I may be affiliated.

[toc] | [next] | [standalone]

#631

On Fri, 2011-09-23, Kevin the Drummer wrote:
> I bought an Intel 1GB network card to replace my 100MB Netgear
> card.  My network speed has DRAMATICALLY decreased.  My network
> monitor shows bursts of ~100KB traffic and then pauses.

When you try to do *what*, exactly? And what does "pause" mean?

> This
> pattern repeats constantly, and between machines in the house,
> not over my DSL.  I searched for info on this issue and found
> several people reported similar experiences over the years.  Many
> also experienced high CPU utilization.  But, my CPU is near 2%
> used.
...
> What might I do to tune up this card for full GB performance?
> Should I just return it and get another card?

It's not clear what your symptoms are. Try to measure using known
traffic.  I think a pingflood between your machine and a nearby one
would be a good first test. You can vary packet sizes etc:

# ping -f -i0 -s1000 somewhere

Also, what does ethtool say? Does it report that you're really running
1Gbit full duplex?

/Jorgen

-- 
  // Jorgen Grahn <grahn@  Oo  o.   .     .
\X/     snipabacken.se>   O  o   .

[toc] | [prev] | [next] | [standalone]

#634

Jorgen Grahn <grahn+nntp@snipabacken.se> wrote:

>  On Fri, 2011-09-23, Kevin the Drummer wrote:
> > I bought an Intel 1GB network card to replace my 100MB Netgear
> > card.  My network speed has DRAMATICALLY decreased.  My network
> > monitor shows bursts of ~100KB traffic and then pauses.
> 
>  When you try to do *what*, exactly? And what does "pause" mean?

The network meter for eth4 in gkrellm alternates between traffic
around 90k and 0, then 80k, then 0, then 73k, then 0, then 78k,
etc...  The state changes every second on the meter.  I think
that timing is special to the measurement method, and not the
traffic.  But?  When I had my 100Mb/s card in the machine, then
even my 1.5Mb/s DSL traffic would reach 158K, sometimes even
above 160K.  I can't see why a 1GB card should fare slower than a
DSL line.

>  It's not clear what your symptoms are. Try to measure using known
>  traffic.  I think a pingflood between your machine and a nearby one
>  would be a good first test. You can vary packet sizes etc:
> 
>  # ping -f -i0 -s1000 somewhere

I measured the traffic while transfering one Linux DVD ISO of
3.7G from machine A in my house to machine B in my house, while
at the same time transfering a vacation video ISO of 1.8G from
machine B in my house to machine A in my house.  The measurements
happened to be done on machine A, which is also the machine
that's having trouble.  All the gear between the two machines is
1Gb/s or better rated/spec'd.

>  Also, what does ethtool say? Does it report that you're really running
>  1Gbit full duplex?

Settings for eth4:
        Supported ports: [ TP ]
        Supported link modes:   10baseT/Half 10baseT/Full 
                                100baseT/Half 100baseT/Full 
                                1000baseT/Full 
        Supports auto-negotiation: Yes
        Advertised link modes:  10baseT/Half 10baseT/Full 
                                100baseT/Half 100baseT/Full 
                                1000baseT/Full 
        Advertised auto-negotiation: Yes
        Speed: 1000Mb/s
        Duplex: Full
        Port: Twisted Pair
        PHYAD: 0
        Transceiver: internal
        Auto-negotiation: on
        Supports Wake-on: umbg
        Wake-on: g
        Current message level: 0x00000007 (7)
        Link detected: yes

I hope the added data is useful.

For what it's worth, I grabbed new kernel source for the e1000
module from Intel last night.  I built and installed that, but
I haven't rebooted yet, as I haven't had time to do that in a
controlled manner, and leave enough time to repair things should
anything go wrong.

Thanks....

-- 
PLEASE post a SUMMARY of the answer(s) to your question(s)!
Unless otherwise noted, the statements herein reflect my personal
opinions and not those of any organization with which I may be affiliated.

[toc] | [prev] | [next] | [standalone]

#641

On Fri, 2011-09-23, Kevin the Drummer wrote:
> Jorgen Grahn <grahn+nntp@snipabacken.se> wrote:
>
>>  On Fri, 2011-09-23, Kevin the Drummer wrote:
>> > I bought an Intel 1GB network card to replace my 100MB Netgear
>> > card.  My network speed has DRAMATICALLY decreased.  My network
>> > monitor shows bursts of ~100KB traffic and then pauses.
>> 
>>  When you try to do *what*, exactly? And what does "pause" mean?
>
> The network meter for eth4 in gkrellm alternates between traffic
> around 90k and 0, then 80k, then 0, then 73k, then 0, then 78k,
> etc...  The state changes every second on the meter.  I think
> that timing is special to the measurement method, and not the
> traffic.  But?  When I had my 100Mb/s card in the machine, then
> even my 1.5Mb/s DSL traffic would reach 158K, sometimes even
> above 160K.  I can't see why a 1GB card should fare slower than a
> DSL line.
>
>>  It's not clear what your symptoms are. Try to measure using known
>>  traffic.  I think a pingflood between your machine and a nearby one
>>  would be a good first test. You can vary packet sizes etc:
>> 
>>  # ping -f -i0 -s1000 somewhere
>
> I measured the traffic while transfering one Linux DVD ISO of
> 3.7G from machine A in my house to machine B in my house, while
> at the same time transfering a vacation video ISO of 1.8G from
> machine B in my house to machine A in my house.  The measurements
> happened to be done on machine A, which is also the machine
> that's having trouble. 

Ah, yes, that's a decent first test.  But I still recommend something
like the ping test to
- eliminate disk read/write speeds from the equation
- get a better momentary view of the losses (gkrellm seems to indicate
  some kind of problem which varies with time, but can't tell you the
  details
- try out lower rates, smaller packets etc to see under which
  situations the problem exists

> All the gear between the two machines is
> 1Gb/s or better rated/spec'd.
>
>>  Also, what does ethtool say? Does it report that you're really running
>>  1Gbit full duplex?
>
> Settings for eth4:
...
>         Speed: 1000Mb/s
>         Duplex: Full
...
>         Link detected: yes
>
> I hope the added data is useful.

I should add here that I'm not good at Ethernet debugging; I'm just
suggesting ways to find out more. Hopefully others here can draw
conclusions from the data.

BR,
Jorgen

-- 
  // Jorgen Grahn <grahn@  Oo  o.   .     .
\X/     snipabacken.se>   O  o   .

[toc] | [prev] | [next] | [standalone]

#646

Jorgen Grahn <grahn+nntp@snipabacken.se> wrote:
> >>  It's not clear what your symptoms are. Try to measure using known
> >>  traffic.  I think a pingflood between your machine and a nearby one
> >>  would be a good first test. You can vary packet sizes etc:
> >> 
> >>  # ping -f -i0 -s1000 somewhere
> >
> > I measured the traffic while transfering one Linux DVD ISO of
> > 3.7G from machine A in my house to machine B in my house, while
> > at the same time transfering a vacation video ISO of 1.8G from
> > machine B in my house to machine A in my house.  The measurements
> > happened to be done on machine A, which is also the machine
> > that's having trouble. 
> 
>  Ah, yes, that's a decent first test.

Just enough to know that there is a problem.

>  But I still recommend something
>  like the ping test to
>  - eliminate disk read/write speeds from the equation
>  - get a better momentary view of the losses (gkrellm seems to indicate
>    some kind of problem which varies with time, but can't tell you the
>    details
>  - try out lower rates, smaller packets etc to see under which
>    situations the problem exists

Good ideas.

> >>  Also, what does ethtool say? Does it report that you're really running
> >>  1Gbit full duplex?
> >
> > Settings for eth4:
>  ...
> >         Speed: 1000Mb/s
> >         Duplex: Full
>  ...
> >         Link detected: yes
> >
> > I hope the added data is useful.
> 
>  I should add here that I'm not good at Ethernet debugging; I'm just
>  suggesting ways to find out more. Hopefully others here can draw
>  conclusions from the data.

OK.  Thanks for giving me things to try!

-- 
PLEASE post a SUMMARY of the answer(s) to your question(s)!
Unless otherwise noted, the statements herein reflect my personal
opinions and not those of any organization with which I may be affiliated.

[toc] | [prev] | [next] | [standalone]

#649

On Sat, 24 Sep 2011 19:30:42 -0500, Kevin the Drummer rearranged some
electrons to say:

>
>> > Settings for eth4:
>>  ...
>> >         Speed: 1000Mb/s Duplex: Full

Try setting to half duplex and see what the result is.

[toc] | [prev] | [next] | [standalone]

#653

On 09/25/2011 01:48 PM, david wrote:
> On Sat, 24 Sep 2011 19:30:42 -0500, Kevin the Drummer rearranged some
> electrons to say:
> 
>>
>>>> Settings for eth4:
>>>  ...
>>>>         Speed: 1000Mb/s Duplex: Full
> 
> Try setting to half duplex and see what the result is.

It's nearly impossible to achieve Gigabit Ethernet with switch.

[toc] | [prev] | [next] | [standalone]

#654

On Tue, 2011-09-27, ein wrote:
> On 09/25/2011 01:48 PM, david wrote:
>> On Sat, 24 Sep 2011 19:30:42 -0500, Kevin the Drummer rearranged some
>> electrons to say:
>> 
>>>
>>>>> Settings for eth4:
>>>>  ...
>>>>>         Speed: 1000Mb/s Duplex: Full
>> 
>> Try setting to half duplex and see what the result is.
>
> It's nearly impossible to achieve Gigabit Ethernet with switch.

Isn't it impossible in practice no matter what's between the NICs?

Anyway, I don't think the OP stated that as his goal.

/Jorgen

-- 
  // Jorgen Grahn <grahn@  Oo  o.   .     .
\X/     snipabacken.se>   O  o   .

[toc] | [prev] | [next] | [standalone]

#656

On 09/27/2011 11:58 PM, Jorgen Grahn wrote:
> On Tue, 2011-09-27, ein wrote:
>> On 09/25/2011 01:48 PM, david wrote:
>>> On Sat, 24 Sep 2011 19:30:42 -0500, Kevin the Drummer rearranged some
>>> electrons to say:
>>>
>>>>
>>>>>> Settings for eth4:
>>>>>  ...
>>>>>>         Speed: 1000Mb/s Duplex: Full
>>>
>>> Try setting to half duplex and see what the result is.
>>
>> It's nearly impossible to achieve Gigabit Ethernet with switch.
> 
> Isn't it impossible in practice no matter what's between the NICs?

I meant: it's neatly impossible to achieve half duplex Gigabit Ethernet
with switch. Sorry.

[toc] | [prev] | [next] | [standalone]

#645

On Sep 22, 10:50 pm, Kevin the Drummer <nob...@cosgroves.us> wrote:

> I bought an Intel 1GB network card to replace my 100MB Netgear
> card.  My network speed has DRAMATICALLY decreased.  My network
> monitor shows bursts of ~100KB traffic and then pauses.  This
> pattern repeats constantly, and between machines in the house,
> not over my DSL.  I searched for info on this issue and found
> several people reported similar experiences over the years.  Many
> also experienced high CPU utilization.  But, my CPU is near 2%
> used.

In my experience, this problem is almost always due to bad cabling.
The most common problems are:

1) Cable that is too long, more than 328 feet.

2) Use of cable or connectors that are not rated for Gigabit ethernet.

3) Improper use of adapters, wiring panels, or splices.

4) Improper assignment of wires to pairs. (You'd be surprised how many
people think that you can just connect corresponding pins of the
connector to wires and have it work correctly.)

5) Pins not used in fast ethernet left unconnected or marginally
connected.

6) Too much wire untwisted at connectors or junctions.

DS

[toc] | [prev] | [next] | [standalone]

#647

David Schwartz <davids@webmaster.com> wrote:
>  On Sep 22, 10:50 pm, Kevin the Drummer <nob...@cosgroves.us> wrote:
> 
> > I bought an Intel 1GB network card to replace my 100MB Netgear
> > card.  My network speed has DRAMATICALLY decreased.  My network
> > monitor shows bursts of ~100KB traffic and then pauses.  This
> > pattern repeats constantly, and between machines in the house,
> > not over my DSL.  I searched for info on this issue and found
> > several people reported similar experiences over the years.  Many
> > also experienced high CPU utilization.  But, my CPU is near 2%
> > used.
> 
>  In my experience, this problem is almost always due to bad cabling.
>  The most common problems are:
> 
>  1) Cable that is too long, more than 328 feet.

I'm using about 3 meters.

>  2) Use of cable or connectors that are not rated for Gigabit ethernet.

That's possible.

>  3) Improper use of adapters, wiring panels, or splices.

If that's the case, then it's that way from the factory.

>  4) Improper assignment of wires to pairs. (You'd be surprised how many
>  people think that you can just connect corresponding pins of the
>  connector to wires and have it work correctly.)

Again.  Factory stuff.

>  5) Pins not used in fast ethernet left unconnected or marginally
>  connected.

You mean in the Netgear router?  Or, in the Intel card?  Or, in the
factory built cable?

>  6) Too much wire untwisted at connectors or junctions.

I'm hoping that falling back on factory built cable is OK.  I have more
cable, and I can swap some in/out to see if I have a bad piece or two.

Thanks...

-- 
PLEASE post a SUMMARY of the answer(s) to your question(s)!
Unless otherwise noted, the statements herein reflect my personal
opinions and not those of any organization with which I may be affiliated.

[toc] | [prev] | [next] | [standalone]

#655

Kevin the Drummer <nobody@cosgroves.us> wrote:
>  David Schwartz <davids@webmaster.com> wrote:
> >  In my experience, this problem is almost always due to bad cabling.
> >  The most common problems are:
> 
> >  2) Use of cable or connectors that are not rated for Gigabit ethernet.
> 
>  That's possible.

I've been trying speed tests between various computers in my
house, isolating sections of cable, switches and NICs in the
process.  I found 3 pieces of CAT5 cable in the process, one of
them 2m long and two others 6m long.  All has been replaced with
CAT5e or CAT6.

I think it was Jorgen who suggested I try:

    ping -f -i0 -s1000 somewhere

I'm not completely sure why, but this doesn't generate as much
net traffic as a file transfer across NFS.  The latter generates
6-8 times more traffic.  I do remember noting that my Netgear
stuff has some automatic DNS attack throttling abilities.  Maybe
that's involved.

As for my NFS file transfers, over the cable sets that are all
CAT5e or better, I now see 226Mb/s on average, simultaneous and
bidirectional.  My limit might be my harddrive.  I'm certainly
better than the 100Mb/s former network, not 10X better, but it's
enough to notice the improvement.  I still have a couple long
CAT5 cables in the network, but at least those are stubs off to
lesser used computers, including a WinXP box.  I'll replace those
at some point.

Thanks all....

-- 
PLEASE post a SUMMARY of the answer(s) to your question(s)!
Unless otherwise noted, the statements herein reflect my personal
opinions and not those of any organization with which I may be affiliated.

[toc] | [prev] | [next] | [standalone]

#657

On 09/28/2011 06:40 AM, Kevin the Drummer wrote:
> Kevin the Drummer <nobody@cosgroves.us> wrote:
>>  David Schwartz <davids@webmaster.com> wrote:
>>>  In my experience, this problem is almost always due to bad cabling.
>>>  The most common problems are:
>>
>>>  2) Use of cable or connectors that are not rated for Gigabit ethernet.
>>
>>  That's possible.
> 
> I've been trying speed tests between various computers in my
> house, isolating sections of cable, switches and NICs in the
> process.  I found 3 pieces of CAT5 cable in the process, one of
> them 2m long and two others 6m long.  All has been replaced with
> CAT5e or CAT6.

Can you precise? Do u have 1000Base-T or 1000Base-TX equipment? From my
experience the 1000Base-T is more common. For this standard you can use
Cat5 cable or higher, so it should not be a problem.

[toc] | [prev] | [next] | [standalone]

#660

ein <ein@no.spam> wrote:
>  On 09/28/2011 06:40 AM, Kevin the Drummer wrote:
> > Kevin the Drummer <nobody@cosgroves.us> wrote:
> >>  David Schwartz <davids@webmaster.com> wrote:
> >>>  In my experience, this problem is almost always due to bad cabling.
> >>>  The most common problems are:
> >>
> >>>  2) Use of cable or connectors that are not rated for Gigabit ethernet.
> >>
> >>  That's possible.
> > 
> > I've been trying speed tests between various computers in my
> > house, isolating sections of cable, switches and NICs in the
> > process.  I found 3 pieces of CAT5 cable in the process, one of
> > them 2m long and two others 6m long.  All has been replaced with
> > CAT5e or CAT6.
> 
>  Can you precise? Do u have 1000Base-T or 1000Base-TX equipment? From my
>  experience the 1000Base-T is more common. For this standard you can use
>  Cat5 cable or higher, so it should not be a problem.

This ADSL2+ modem with GB router and dual channel wireless does not
specify that I can tell.  Can you?  I have the User Manual for it, and
it's not specific as to any special type of GB ethernet.

    http://www.newegg.com/Product/Product.aspx?Item=N82E16833122411

This 8 port GB switch says 1000Base-T and 100Base-TX.  Notice that the
"TX" portion is only 100Base-TX.  The "Details" tab at this link calls
out the IEEE 802.3 specifics.

    http://www.newegg.com/Product/Product.aspx?Item=N82E16833122111

Thanks...

-- 
PLEASE post a SUMMARY of the answer(s) to your question(s)!
Unless otherwise noted, the statements herein reflect my personal
opinions and not those of any organization with which I may be affiliated.

[toc] | [prev] | [next] | [standalone]

#648

Kevin the Drummer <nobody@cosgroves.us> wrote:
>  I bought an Intel 1GB network card to replace my 100MB Netgear
>  card.  My network speed has DRAMATICALLY decreased.  My network
>  monitor shows bursts of ~100KB traffic and then pauses.  This
>  pattern repeats constantly, and between machines in the house,
>  not over my DSL.  I searched for info on this issue and found
>  several people reported similar experiences over the years.  Many
>  also experienced high CPU utilization.  But, my CPU is near 2%
>  used.
> 
>  My house is 1GB over CAT6 cables.  All of the networking hardware
>  outside my computer cases is Netgear stuff, switches & routers.
>  The Intel card uses the e1000 kernel module in a 2.6.17.14
>  kernel.  The specific card I'm using is an "Intel PWLA8391GT
>  Desktop Adapter PRO/1000 GT 10/100/1000Mbps PCI"
> 
>  What might I do to tune up this card for full GB performance?
>  Should I just return it and get another card?  I bought Intel
>  thinking that would work, and because it was highly rated on
>  the shopping site I visited.

I visited the Intel website, downloaded a new e1000 driver starting here:

    http://support.intel.com/support/go/linux/e1000.htm

Then I clicked on "Network Connectivity" and "Intel Desktop Adapters"
which allowed this link to appear

http://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&DwnldID=9180&keyword=%22e1000-%22&lang=eng

where I downloaded version 8.0.30 of the e1000 kernel module
source from Intel.  I built and installed that.  After my next
reboot my system no longer exhibits the brutally slow connection,
nor are the 1 second ethernet pauses visible in gkrellm.

I don't know if I have full speed yet, having not tested with
pings yet.  And I haven't tried swapping any cables to weed out
potential bad ones.  But, the speed is easily 100x what it was
before.

Thanks all....

-- 
PLEASE post a SUMMARY of the answer(s) to your question(s)!
Unless otherwise noted, the statements herein reflect my personal
opinions and not those of any organization with which I may be affiliated.

[toc] | [prev] | [standalone]

Back to top | Article view | comp.os.linux.networking

csiph-web