Gigabit Network only connects @ 100Mbit on startup

Hi everybody,

I am having a rather serious networking problem that I hope someone can help
me with.

I have a small network, consisting of an ADSL router with a 100mbit hub, a
Gigabit switch, one machine with a 100mbit NIC and two machines with gigabit
NICs.

The 100mbit machine is directly connected to the router, and the router is
connected to the gigabit switch.
The other two PCs are directly connected to the gigabit switch.
All machines run WinXP Pro, SP2. All cabling is CAT-5, certified for Gigabit
Ethernet.
The router is an SMC Barricade 7404BRA, and the gig switch an SMC EZSwitch
10/100/1000 8505T.

Now the problem I have is weird, and it goes like this : When I first boot
any of the two Gig PCs, they will not connect to the Gig switch. They will
not connect at all, they will just keep trying and keep failing. If I
manually set them to 100Mbit/Full duplex, they instantly connect, but even
then the connection is ridiculously slow. Anything more than 100/full set in
the NICs' settings will fail to even connect at all.

Now, the really weird thing is that after a couple of hours of running time,
I can go to the NICs' settings and either set them to Auto Negotiate or
directly to 1000/full, and they connect fine, with perfect Gig speeds, too -
without changing *anything* else ! But I repeat, that can only happen after
both machines have been on for a while. At that point, I can reboot the PCs,
or even shut them down and turn them on again after a few minutes, and there
is no problem. If I shut them down for a few hours and boot them up again,
the problem appears again.

I 've looked everywhere, and tried anything I could think of. I
reinstalled/updated drivers, replaced all cabling, reset the router and
switch, even changed NICs (tried with the on-board intel Pro/1000 CT ones,
and then disabled those and used 3Com gigabit NICs instead. The same thing
happens. I also eliminated the possibility of a faulty switch, since I
tested it on another network and it worked fine every time.
I should also mention that connecting the two gig PCs directly with a
crosswired cable works perfectly every time too (which points to a faulty
switch, but like I said... I tested it and it worked fine on another setup).

I 'm really confused here... I 've searched all over the place but couldn't
find anything. I would appreciate any help I can get from anyone !

Thanks, and apologies for the kinda long post - I wanted to make the whole
setup and problem clear from the start.

nick

--


Nick Hatzichristos
AnimateThis
nixx /at/ animatethis /dot/ gr

ICQ# : 48636115

Nick Hatzichristos wrote:
> Hi everybody,
>
> I am having a rather serious networking problem that I hope someone can help
> me with.
>
> I have a small network, consisting of an ADSL router with a 100mbit hub, a
> Gigabit switch, one machine with a 100mbit NIC and two machines with gigabit
> NICs.
>
> The 100mbit machine is directly connected to the router, and the router is
> connected to the gigabit switch.
> The other two PCs are directly connected to the gigabit switch.
> All machines run WinXP Pro, SP2. All cabling is CAT-5, certified for Gigabit
> Ethernet.
> The router is an SMC Barricade 7404BRA, and the gig switch an SMC EZSwitch
> 10/100/1000 8505T.
>
> Now the problem I have is weird, and it goes like this : When I first boot
> any of the two Gig PCs, they will not connect to the Gig switch. They will
> not connect at all, they will just keep trying and keep failing. If I
> manually set them to 100Mbit/Full duplex, they instantly connect, but even
> then the connection is ridiculously slow. Anything more than 100/full set in
> the NICs' settings will fail to even connect at all.
>
> Now, the really weird thing is that after a couple of hours of running time,
> I can go to the NICs' settings and either set them to Auto Negotiate or
> directly to 1000/full, and they connect fine, with perfect Gig speeds, too -
> without changing *anything* else ! But I repeat, that can only happen after
> both machines have been on for a while. At that point, I can reboot the PCs,
> or even shut them down and turn them on again after a few minutes, and there
> is no problem. If I shut them down for a few hours and boot them up again,
> the problem appears again.
>
> I 've looked everywhere, and tried anything I could think of. I
> reinstalled/updated drivers, replaced all cabling, reset the router and
> switch, even changed NICs (tried with the on-board intel Pro/1000 CT ones,
> and then disabled those and used 3Com gigabit NICs instead. The same thing
> happens. I also eliminated the possibility of a faulty switch, since I
> tested it on another network and it worked fine every time.
> I should also mention that connecting the two gig PCs directly with a
> crosswired cable works perfectly every time too (which points to a faulty
> switch, but like I said... I tested it and it worked fine on another setup).
>
> I 'm really confused here... I 've searched all over the place but couldn't
> find anything. I would appreciate any help I can get from anyone !
>
> Thanks, and apologies for the kinda long post - I wanted to make the whole
> setup and problem clear from the start.
>
> nick
>

That _is_ weird. I'll start the brainstorming:

It sounds like something thermal; that you've swapped NICS with no
change suggests it's the switch. Might the other network in which
you tested the switch be environmentally different (either in
temperature or perhaps line voltage)? Also, FWIW, my impression is
that switches heat up as they are actually called upon to perform
their function (more than just being powered up).

Out of curiosity, when you say you replaced all the cables, does that
include the drop cables? Since we're plumbing the depths of the weird,
it occurs to me that a bimetallic interaction at the drop cable
connections could generate a small bias voltage that might be
temperature sensitive, and might tip the balance in a marginal system.

In the mode where they won't connect at all, do the status lights on
the NICs and switch reveal anything?

How long are the cable runs?

Clearly one thing you could try is replacing the switch (at least
temporarily) and see whether the problem disappears.

--
The e-mail address in our reply-to line is reversed in an attempt to
minimize spam. Our true address is of the form che...@prodigy.net.

Hi CJT, thanks for replying and brainstorming with me on this... I 'll try
to clear some things more.


> That _is_ weird. I'll start the brainstorming:

Good, at least I haven't got mad or anything... : )


> It sounds like something thermal; that you've swapped NICS with no
> change suggests it's the switch. Might the other network in which
> you tested the switch be environmentally different (either in
> temperature or perhaps line voltage)?

Yes, it was a completely different environment actually - line voltage was
most probably the same, but everything else could have been different.


> Also, FWIW, my impression is
> that switches heat up as they are actually called upon to perform
> their function (more than just being powered up).

I see. But still, it should work right away, shouldn't it ?


> Out of curiosity, when you say you replaced all the cables, does that
> include the drop cables? Since we're plumbing the depths of the weird,
> it occurs to me that a bimetallic interaction at the drop cable
> connections could generate a small bias voltage that might be
> temperature sensitive, and might tip the balance in a marginal system.

I see what you mean, but I don't think I get what you mean by "drop cables"
(sorry). What I meant by "all cables" was all PC<>switch CAT5 cables,
PC<>router and router<>switch. The only thing I haven't tried replacing, now
that I think of it, is the ADSL line cable from the wall outlet to the
router, but this is completely irrelevant, isn't it ?


> In the mode where they won't connect at all, do the status lights on
> the NICs and switch reveal anything?

They light up (indicating 1gig connections) for a second or two, while the
PCs try to connect, and they turn off when the connection fails. And then
they light up again, then turn off, and so on. No troubleshooting signal I
'm afraid (checked the manual, too).


> How long are the cable runs?

100mbit PC > Router, 6-7 feet (~2-2,5m)
Router > Switch, around 45 feet (~15m, that's the longest one)
"Problematic" gig PCs > Switch, 10-12 feet (~3-4m) each.


> Clearly one thing you could try is replacing the switch (at least
> temporarily) and see whether the problem disappears.

I think I will try that anyway... I meant to, but abandoned the idea when I
saw the very same piece work perfectly on another network.

One other thing I just realised, is that when I took the switch for testing,
it could have been after it had already worked for a few hours (I can't
remember, but it could). That would point to your heat-up theory (since it
may have been already heated up when it was tested), but what if that theory
is correct ? Would it mean that the switch is actually defective or
something ? That can't be the "right" way a switch works, can it ?

Thank you for trying to tackle this with me, I appreciate your help a lot

nick

--


Nick Hatzichristos
AnimateThis
nixx /at/ animatethis /dot/ gr

ICQ# : 48636115

Nick Hatzichristos wrote:
> Hi CJT, thanks for replying and brainstorming with me on this... I 'll try
> to clear some things more.
>
>
>
>>That _is_ weird. I'll start the brainstorming:
>
>
> Good, at least I haven't got mad or anything... : )
>
>
>
>>It sounds like something thermal; that you've swapped NICS with no
>>change suggests it's the switch. Might the other network in which
>>you tested the switch be environmentally different (either in
>>temperature or perhaps line voltage)?
>
>
> Yes, it was a completely different environment actually - line voltage was
> most probably the same, but everything else could have been different.
>
>
>
>>Also, FWIW, my impression is
>>that switches heat up as they are actually called upon to perform
>>their function (more than just being powered up).
>
>
> I see. But still, it should work right away, shouldn't it ?

It should. Then again, the whole system _should_ work. <g>
Some electronics tend to work better when warmed up. Some NICs
are more tolerant of variances than others, too.

>
>
>
>>Out of curiosity, when you say you replaced all the cables, does that
>>include the drop cables? Since we're plumbing the depths of the weird,
>>it occurs to me that a bimetallic interaction at the drop cable
>>connections could generate a small bias voltage that might be
>>temperature sensitive, and might tip the balance in a marginal system.
>
>
> I see what you mean, but I don't think I get what you mean by "drop cables"
> (sorry). What I meant by "all cables" was all PC<>switch CAT5 cables,
> PC<>router and router<>switch.

[I might not be envisioning this quite like it is. To me, a drop cable
is one from, e.g., a PC to the wall, or from the wall to the switch.
Then there's the in-wall wiring. That you said you had certified the
cables suggested to me that there was some in-wall wiring involved,
since normally people don't certify drop cables on-site. What you just
wrote suggests I might not be understanding completely.]

(As I read further, I realized this [] is probably irrelevant, anyway,
since you've apparently just got short cables connecting the offending
PCs to the switch)

BTW, what's the nature of the certification that was done?

The only thing I haven't tried replacing, now
> that I think of it, is the ADSL line cable from the wall outlet to the
> router, but this is completely irrelevant, isn't it ?
>
I wouldn't worry about that one.

>
>
>>In the mode where they won't connect at all, do the status lights on
>>the NICs and switch reveal anything?
>
>
> They light up (indicating 1gig connections) for a second or two, while the
> PCs try to connect, and they turn off when the connection fails. And then
> they light up again, then turn off, and so on. No troubleshooting signal I
> 'm afraid (checked the manual, too).
>
I don't claim to be an expert on this particular aspect, but that sounds
to me like a failure negotiating. Is the switch one that auto-detects
both speed and the need for a crossover connection? If so, can it be
forced to non-crossover mode (i.e. skip the X vs. II determination)?
>
>
>>How long are the cable runs?
>
>
> 100mbit PC > Router, 6-7 feet (~2-2,5m)
> Router > Switch, around 45 feet (~15m, that's the longest one)
> "Problematic" gig PCs > Switch, 10-12 feet (~3-4m) each.
>
It seems to me that for the problem described, just that last bit
counts, since presumably you could disconnect the router from the
switch and things would be the same on the LAN.

Are those 3-4m cables "store bought?" Or were they fabricated on-site?
I'm always suspicious of cabling, because so many people get it wrong.
Out of curiosity, if they were fabricated on-site, to which standard
do they adhere -- 568A or 568B?

>
>
>>Clearly one thing you could try is replacing the switch (at least
>>temporarily) and see whether the problem disappears.
>
>
> I think I will try that anyway... I meant to, but abandoned the idea when I
> saw the very same piece work perfectly on another network.
>
> One other thing I just realised, is that when I took the switch for testing,
> it could have been after it had already worked for a few hours (I can't
> remember, but it could). That would point to your heat-up theory (since it
> may have been already heated up when it was tested), but what if that theory
> is correct ? Would it mean that the switch is actually defective or
> something ? That can't be the "right" way a switch works, can it ?

They should work cold, but that's no guarantee this particular one does.
Even a switch that meets specs could have intermittent problems if
something else (e.g. cabling) is marginal.
>
> Thank you for trying to tackle this with me, I appreciate your help a lot

Merry Christmas. :-)
>
> nick
>


--
The e-mail address in our reply-to line is reversed in an attempt to
minimize spam. Our true address is of the form che...@prodigy.net.

"CJT" <(E-Mail Removed)> wrote in message
news:(E-Mail Removed)...

> [I might not be envisioning this quite like it is...SNIP]
> ...
> ...
> BTW, what's the nature of the certification that was done?

I probably mislead you about the certification - all cables are store
bought, no in-wall wiring, and by "certified" I meant that they all have
"gigabit ethernet certified" printed across
No on-site certification, and certainly no on-site fabrication of wires...


> I don't claim to be an expert on this particular aspect, but that sounds
> to me like a failure negotiating. Is the switch one that auto-detects
> both speed and the need for a crossover connection? If so, can it be
> forced to non-crossover mode (i.e. skip the X vs. II determination)?

It auto-detects speed, but only works with II cables (tried crossover and it
failed, and that was in "working OK" mode).


> It seems to me that for the problem described, just that last bit
> counts, since presumably you could disconnect the router from the
> switch and things would be the same on the LAN.

Exactly (I tried that too). I think it's clear that my problem is with the
gig PCs <> switch connections.


> Are those 3-4m cables "store bought?" Or were they fabricated on-site?
> I'm always suspicious of cabling, because so many people get it wrong.
> Out of curiosity, if they were fabricated on-site, to which standard
> do they adhere -- 568A or 568B?

Like I mentioned above, they're store-bought and they are of the 568B
standard ("568B.2" is printed across, along with other tech specs).


> They should work cold, but that's no guarantee this particular one does.
> Even a switch that meets specs could have intermittent problems if
> something else (e.g. cabling) is marginal.

I 'll do some more hot/cold tests to see if that's actually the case, and
then I 'll see if I can get some other make/model for testing.



> Merry Christmas. :-)

Merry Christmas to you too (and everyone else !) and a happy, speedy,
bottleneck-free new year !

thank you once again

nick

--


Nick Hatzichristos
AnimateThis
nixx /at/ animatethis /dot/ gr

ICQ# : 48636115

Nick Hatzichristos wrote:
> "CJT" <(E-Mail Removed)> wrote in message
> news:(E-Mail Removed)...
>
>
>>[I might not be envisioning this quite like it is...SNIP]
>>...
>>...
>>BTW, what's the nature of the certification that was done?
>
>
> I probably mislead you about the certification - all cables are store
> bought, no in-wall wiring, and by "certified" I meant that they all have
> "gigabit ethernet certified" printed across
> No on-site certification, and certainly no on-site fabrication of wires...
>
>
>
>>I don't claim to be an expert on this particular aspect, but that sounds
>>to me like a failure negotiating. Is the switch one that auto-detects
>>both speed and the need for a crossover connection? If so, can it be
>>forced to non-crossover mode (i.e. skip the X vs. II determination)?
>
>
> It auto-detects speed, but only works with II cables (tried crossover and it
> failed, and that was in "working OK" mode).
>
>
>
>>It seems to me that for the problem described, just that last bit
>>counts, since presumably you could disconnect the router from the
>>switch and things would be the same on the LAN.
>
>
> Exactly (I tried that too). I think it's clear that my problem is with the
> gig PCs <> switch connections.
>
>
>
>>Are those 3-4m cables "store bought?" Or were they fabricated on-site?
>>I'm always suspicious of cabling, because so many people get it wrong.
>>Out of curiosity, if they were fabricated on-site, to which standard
>>do they adhere -- 568A or 568B?
>
>
> Like I mentioned above, they're store-bought and they are of the 568B
> standard ("568B.2" is printed across, along with other tech specs).
>
>
>
>>They should work cold, but that's no guarantee this particular one does.
>>Even a switch that meets specs could have intermittent problems if
>>something else (e.g. cabling) is marginal.
>
>
> I 'll do some more hot/cold tests to see if that's actually the case, and
> then I 'll see if I can get some other make/model for testing.
>
>
>
>
>>Merry Christmas. :-)
>
>
> Merry Christmas to you too (and everyone else !) and a happy, speedy,
> bottleneck-free new year !
>
> thank you once again
>
> nick
>
It sounds like you've done everything right, and it should "just work."

I'm fresh out of ideas right now, and it's almost dinner time in Texas,
so I'll just say "best of luck with it!" Maybe something we've
discussed will trigger some thoughts in other readers.

--
The e-mail address in our reply-to line is reversed in an attempt to
minimize spam. Our true address is of the form che...@prodigy.net.

Nick Hatzichristos wrote:
> Now the problem I have is weird, and it goes like this : When I first boot
> any of the two Gig PCs, they will not connect to the Gig switch. They will
> not connect at all, they will just keep trying and keep failing. If I
> manually set them to 100Mbit/Full duplex, they instantly connect, but even
> then the connection is ridiculously slow. Anything more than 100/full set in
> the NICs' settings will fail to even connect at all.

Now, having glanced through some of the other discussions you've had in
this thread, have you tried fiddling with any of the the other advanced
network card configuration settings available to you? These parameters
are all different depending on adapter, but in my 100BT NICs I also have
settings for Adaptive Interrupt, Flow Control, Network Address, Receive
Buffers, Transmit Buffers, and Validate Packet Length. I've left all of
those at defaults in my case, since everything is working fine. In your
case, you may want to check into them?

Now it's entirely possible that you've run into one of those early
teething issues with new technology. This sort of thing used to occur
with 100BT hardware too in the early days, though these days that
technology has been around so long that we don't remember them. I'm sure
they even occurred in the really old days with 10BT, when people were
switching away from coax Ethernet to twisted-pair, but I was barely
cognizant of networking back in those days.

Yousuf Khan

"Yousuf Khan" <(E-Mail Removed)> wrote in message
news:CuidnQrKFMWUmVPcRVn-(E-Mail Removed)...

> Now, having glanced through some of the other discussions you've had in
> this thread, have you tried fiddling with any of the the other advanced
> network card configuration settings available to you? These parameters
> are all different depending on adapter, but in my 100BT NICs I also have
> settings for Adaptive Interrupt, Flow Control, Network Address, Receive
> Buffers, Transmit Buffers, and Validate Packet Length. I've left all of
> those at defaults in my case, since everything is working fine. In your
> case, you may want to check into them?


Hi Yousuf, thanks for replying

There are, of course, advanced settings in the NIC's control panel, and I
have tried changing a couple of them. But seeing as I don't know the exact
effect of each setting, and given that the possible combinations are
endless, I haven't tried playing with more than just a couple - you know, I
could make things worse
Plus, the fact that after a while the whole thing settles down and works as
expected is a good indication that none of those settings need fiddling
with...

I 'll keep looking into this, I really hope I find what's wrong - it's
frustrating, to say the least, to have to wait a few hours every morning
before I get my LAN working

Anyway, thanks again for sharing your idea : ) Merry Christmas !

nick

--


Nick Hatzichristos
AnimateThis
nixx /at/ animatethis /dot/ gr

ICQ# : 48636115

Nick Hatzichristos wrote:
> "Yousuf Khan" <(E-Mail Removed)> wrote in message
> news:CuidnQrKFMWUmVPcRVn-(E-Mail Removed)...
>
>
>>Now, having glanced through some of the other discussions you've had in
>>this thread, have you tried fiddling with any of the the other advanced
>>network card configuration settings available to you? These parameters
>>are all different depending on adapter, but in my 100BT NICs I also have
>>settings for Adaptive Interrupt, Flow Control, Network Address, Receive
>>Buffers, Transmit Buffers, and Validate Packet Length. I've left all of
>>those at defaults in my case, since everything is working fine. In your
>>case, you may want to check into them?
>
>
>
> Hi Yousuf, thanks for replying
>
> There are, of course, advanced settings in the NIC's control panel, and I
> have tried changing a couple of them. But seeing as I don't know the exact
> effect of each setting, and given that the possible combinations are
> endless, I haven't tried playing with more than just a couple - you know, I
> could make things worse
> Plus, the fact that after a while the whole thing settles down and works as
> expected is a good indication that none of those settings need fiddling
> with...
>
> I 'll keep looking into this, I really hope I find what's wrong - it's
> frustrating, to say the least, to have to wait a few hours every morning
> before I get my LAN working
>
> Anyway, thanks again for sharing your idea : ) Merry Christmas !
>
> nick
>

It might be worth checking that you have the most recent firmware in
the switch:

http://www.smc.com/index.cfm?event=v...scid=&pid=1138

Interestingly, that page indicates it does auto MDI-I/X.

--
The e-mail address in our reply-to line is reversed in an attempt to
minimize spam. Our true address is of the form che...@prodigy.net.

For the record :

I ended up replacing the SMC switch with a different one (a 3com). It worked
perfectly right away, and still works perfectly on the third day of testing,
right from the start (no need to "warm up" !).

The really weird thing is, however, that the SMC switch still worked
perfectly in a different environment than my own. It seems that whatever it
was that was affecting it, the 3Com one is immune to

I still can't explain it, but the problem is gone now so I think I 'll just
seal this and archive it away in my "X-files" drawer : )

Big thanks to all who tried to help me get to the bottom of this.

Happy new year !

nick

--


Nick Hatzichristos
AnimateThis
nixx /at/ animatethis /dot/ gr

ICQ# : 48636115