Web and Unmanaged
1748209 Members
2805 Online
108759 Solutions
New Discussion

1Gb/s links freezing up instead of gracefully falling back to 100Mb/s

 
BuGless
Occasional Advisor

1Gb/s links freezing up instead of gracefully falling back to 100Mb/s

I have a setup with five HP1810-24Gv2 and three HP1810-8G switches.

This used to be a setup with 1Gb/s Zyxel switches.

 

When I replace the Zyxels with the HP1810s, I notice the following:

 

a. Some links don't come up.  I.e. HP1810 at either side configured in Auto-negotiate-speed mode, the Zyxel happily did 1Gb/s across, the HPs simply get stuck, the link light comes up once every 5 seconds or so, and dies immediately again.  When I configure one side for 100Mb/s, it works.  *If* I exchange either one of the sides with a Zyxel (i.e. Zyxel auto to HP auto), it happily does 1Gb/s without hiccups.

 

b. Some links *do* come up at 1Gb/s, but experience intermittent transmission lapses (retrains?) of up to 8 seconds, happening randomly at intervals between every 30 to 300 seconds.  Setting *these* links to 100Mb/s at either side simply causes a lockup of the port.  Setting these links to 100Mb/s at both sides results in a lockup.  Same for 10Mb/s (FDX or HDX, does not matter).  The only thing that makes them work is setting them to Auto at both ends, and then they negotiate 1000Mb/s (but with intermittent carrier loss).

 

I'm a bit surprised (to say the least), that HP switches have trouble talking to HP switches, but have less problems talking to Zyxels.  I would have expected better quality from HP and find it quite puzzling that the negotiation handshake code in the HP switch firmware is so bugridden as to be unable to talk to itself under certain circumstances.

 

Any pointers or firmware updates are welcome.

 

 

P.S. Thisthread ahs been moevd from ProCurve / ProVision-Based to Web and Unmanaged. - Hp Forum Moderator

8 REPLIES 8
Vince-Whirlwind
Honored Contributor

Re: 1Gb/s links freezing up instead of gracefully falling back to 100Mb/s

a. is exactly what you get with dud cabling that has 1236 correctly wired but doesn't have all 4 good pairs. Are you using Cat6 cables?

 

b. also sounds like a cable quality issue.

 

I would assume you are using the same cabling when you put the older switches on, so putting the bad cabling issue aside for the moment, could it be something else?

 

What is the configuration difference between the old and the new? Do the old switches have broadcast storm suppression or STP configured and the new ones do not? Maybe check the logs on the switches for clues.

 

Blaming bugs in the switches (even cheap switches like the 1810s) is probably not the first conclusion to jump to...

BuGless
Occasional Advisor

Re: 1Gb/s links freezing up instead of gracefully falling back to 100Mb/s

> a. is exactly what you get with dud cabling that has 1236 correctly wired but doesn't have all 4 good pairs. Are you using Cat6 cables?

 

I'm using CAT6/7 cables of about 30m length, but at both ends there is a 2m CAT5 extension that actually goes into the switch, so, yes, the cabling is less than stellar.


> b. also sounds like a cable quality issue.

 

Well, yes, it obviously is related to the cable quality, because this problem does not occur on short cables between the switches (<2m), where autonegotiation and link stability is perfect.


> I would assume you are using the same cabling when you put the older switches on, so putting the bad cabling issue aside for the moment, could it be something else?

 

I indeed kept the cabling identical to the previous setup using the old switches.  Given the fact that the older switches *did* do flawless autonegotiation and didn't have hiccups at 1Gb/s using the exact same cabling, *and* that even a HP1810 paired with an old switch autonegotiates perfectly and flawless over these cables, makes me a bit suspicious that two identical HP1810s have great difficulty to both auto and/or forcibly negotiate and maintaining a stable connection.


> What is the configuration difference between the old and the new? Do the old switches have broadcast storm suppression or STP configured and the new ones do not? Maybe check the logs on the switches for clues.

 

The old switches did not have STP, the new ones do, however there are no redundant loops in the network.  Broadcast storm suppression is off on both.  In any case, the problem is not that the switch refuses to forward, it's that the port/link status refuses to come or stay up.  Thank you for the hint to check the logs; however it does not show more information (I just checked) than that the link goes up or down. 

 

> Blaming bugs in the switches (even cheap switches like the 1810s) is probably not the first conclusion to jump to...

 

I realise that the 1810s are at the low end of the performance range, nonetheless I'd expect the difference between the low end and high end switch types to be mostly in CPU and switch performance and capabilities, and to a lesser extent in the signal processing capabilities at port/link level and/or the autonegotiate protocol being run on the ports.  Given that, I'd say that the preliminary evidence most likely points in the direction of some flawed autonegotiation logic being employed on the ports of the HP1810G switches.

Vince-Whirlwind
Honored Contributor

Re: 1Gb/s links freezing up instead of gracefully falling back to 100Mb/s

Am I understanding this correctly?

 

a.

"I'm using CAT6/7 cables of about 30m length, but at both ends there is a 2m CAT5 extension that actually goes into the switch"

 

b. 

"this problem does not occur on short cables between the switches (<2m), where autonegotiation and link stability is perfect."

 

You seem to have documented cable quality issues, and your symptoms are exactly those you get with cable quality issues.

Having found the issue, I don't really understand why you are talking about "buggy switches".

 

"the older switches *did* do flawless autonegotiation and didn't have hiccups at 1Gb/s using the exact same cabling,"

 

Can you check the exact make and model of these older switches for us?

 

 

BuGless
Occasional Advisor

Re: 1Gb/s links freezing up instead of gracefully falling back to 100Mb/s

> You seem to have documented cable quality issues, and your symptoms are exactly those you get with cable quality issues.
> Having found the issue, I don't really understand why you are talking about "buggy switches".

 

Given perfect conditions, equipment is more likely to have stellar performance.  Equipment can really show what it is worth given less than ideal circumstances.  The cutoff point where the performance goes from stellar to less determines the engineering quality, since we're dealing with analogue signal processing, this is a sliding scale.  I'd expect HP to either match or exceed the quality of Zyxel given similar circumstances/cable quality.

 

To clarify, what I would have expected is something like this:

 

Zyxel 1528 <-> Zyxel 1528: Works and negotiates on the bad cabling, does all speeds 1Gb/s and 100Mb/s and will degrade to 100Mb/s either forced or automatic (the latter if cable quality becomes even worse).

HP1810G <-> HP1810G: Should be identical (or better) behaviour and perfomance as the Zyxel<->Zyxel connection.

HP1810G <-> Zyxel 1528: Could fail in any number of ways due to implementation differences in the autonegotiation protocol which could cause the autonegotiation to fail completely and you'd end up with a 100Mb/s or dead link.

 

However, what I get is:

Zyxel <-> Zyxel: as described above.

HP1810G <-> HP1810G: Autonegotiation on both sides sometimes hangs resulting in a dead link (no fallback to 100Mb/s), or autonegotiation *only* works if both sides are set to auto but then results in a link which is periodically unstable, and on some links setting both sides to a fixed 100FDX or 10FDX (!) results in a dead link.

HP1810G <-> Zyxel: Autonegotiation on both sides results in a 1Gb/s link which is stable.

 

The above testresults clearly indicate that the engineering team of Zyxel did a *better* job of being electrically compatible with HP than HP did being electrically compatible with *itself*.  That I consider "bad engineering" on the HP side.  The cabling in question is *more* than adequate for 100Mb/s, and in some places probably is not ideal for 1Gb/s, so yes, I do plan to replace the cabling with CAT6/7 end-to-end to improve on this.  Nonetheless I find it worrisome that the engineering tolerance of HP is so much worse than that of Zyxel.

 

Let's assume that the analogue circuitry of the HP is inferior to that of the Zyxel, in that case it would be normal that Zyxel <-> Zyxel would negotiate 1Gb/s and that HP <-> HP would negotiate 100Mb/s and that Zyxel <-> HP would negotiate 100Mb/s (or a dodgy 1Gb/s with intermittent failures because the HP can't keep up at the receiving end).

But, what happens instead is that HP <-> HP utterly fails to negotiate something sensible (like 100Mb/s) and simply leaves the link for dead.  What also happens is that Zyxel <-> HP negotiates a flawless 1Gb/s connection.  So I'd guess that the analogue circuitry of the HP should probably be close in quality to that of the Zyxel, but that the logic in the autonegotiation (even in the forced case) of the HP is not well tested with *itself* under bad CAT6 cabling conditions (and thus buggy).


> "the older switches *did* do flawless autonegotiation and didn't have hiccups at 1Gb/s using the exact same cabling,"

> Can you check the exact make and model of these older switches for us?

 

The old switches are Zyxel ES-1528 switches on port 25 and 26 (Gb/s ports).

BuGless
Occasional Advisor

Re: 1Gb/s links freezing up instead of gracefully falling back to 100Mb/s

Small update: A slight change in configuration of the HP1810 solves a single specific failurecase.  This does not solve all the described failures.

 

Turning off "Low-Traffic Idle (EEE)" seems to eliminate the forced speed settings (turning off autoneg) problem on the link that experiences intermittent transmission lapses.  I turn that off only on one side, and after having done that, the switches allow other settings than Autonegotiate to be specified (e.g. 1000FDX, 100FDX, 10FDX) and things still work.  Previously this was not the case on this link, it *required* Autonegotiate on on both sides, otherwise it would be a dead link.

 

In the same vein, I tried to permutate all other Green-settings, but turning off the other two did not make a difference (at least not on this particular problem).

 

This would indicate at least one obvious bug in the firmware (undesirable interaction between link speed negotiation and the "Low-Traffic Idle (EEE)" feature , but it sadly does not solve all the problems encountered.  Are these forums being monitored by HP so that this turns into a trouble ticket automatically, or how do I create a support trouble ticket for this?

BuGless
Occasional Advisor

Re: 1Gb/s links freezing up instead of gracefully falling back to 100Mb/s

With regard to opening a support ticket; we do not have any fancy support contracts with HP, we're just a customer trying to assist in getting the product bugfree.

Vince-Whirlwind
Honored Contributor

Re: 1Gb/s links freezing up instead of gracefully falling back to 100Mb/s

You say you expect the following:

"HP1810G <-> HP1810G: Autonegotiation on both sides sometimes hangs resulting in a dead link (no fallback to 100Mb/s)"

 

If you check the IEEE standard for 1Gb, you will see that

 - auto/auto is the standard, and

 - there is no such thing as "fallback to 100Mb/s"

 

The behaviour you describe is precisely the behaviour all hardware demonstrates when trying to autonegotiate 1Gb over faulty cabling: the auto-neg is carried out using wires 1236 ONLY, and if the remaining 4 wires are not 100%, the link fails following autonegotiation. This is standard behaviour.

 

The behaviour you say you get from your 100Mb Zyxel switch is exactly what I have seen many times over when people are upgrading from a 100Mb switch to a 1Gb switch.

Sub-standard cabling that gave no problem at 100Mb causes problems with 1Gb switches.

 

The fact you say you were actually using the 1Gb ports on your Zyxel is the surprising bit to me, because the behaviour you describe does not match any  use of 1Gb I have ever seen, and does not match the IEEE standard, either.

Either Zyxel has a unique non-IEEE-standard 1Gb implementation, or you were using 100Mb ports, not 1Gb.

BuGless
Occasional Advisor

Re: 1Gb/s links freezing up instead of gracefully falling back to 100Mb/s

> If you check the IEEE standard for 1Gb, you will see that

> - auto/auto is the standard, and

> - there is no such thing as "fallback to 100Mb/s"

 

Ok, I admit I assumed that's what should happen, because that's the behaviour I've seen so far on all the switches I've been using.  I.e. plug in the cables, notice that they negotiate only 100Mb/s, then improve the cabling, try again, and notice that they now do 1Gb/s.  I actually checked the standard this time, and you seem to be right, a fallback to 100Mb/s is not in there.

 

> The fact you say you were actually using the 1Gb ports on your Zyxel is the surprising bit to me, because the behaviour you describe does not match any  use of 1Gb I have ever seen, and does not match the IEEE standard, either.

Either Zyxel has a unique non-IEEE-standard 1Gb implementation, or you were using 100Mb ports, not 1Gb.

 

 

Well, let me confirm once more that everything I wrote and rapported above is correct.  Yes, that includes the usage of the 1Gb/s ports on the Zyxel.  We're talking about an inter-switch network based on 1Gb/s links.  The connections from any of those switches to local computerequipment is short enough in all cases to support flawless 1Gb/s speeds.  The whole slew of problems only starts to arise with the inter-switch 1Gb/s links.  And it still is so that as soon as I replace only one side of the problematic connections with a Zyxel switch on the 1Gb/s port, that then the HP will flawlessly negotiate 1Gb/s and works just fine on that connection from then on.