Drops RX on a 6km fiber connection

[Ronni Jorgensen]

Hello,

I have 2 x HP6108 swiches with 2 x mini gbic 1000mbit fiber cards. The line betwin the 2 switches are 6,2km long. An the fiber type is Single mode fiber.

The 2 ports is set to "Auto" (have tried with 1000FDx but it is the same)

On one of these switches I got about 1-200 Drops Rx over the day. But switch2 I showing no drops at all.

The servers in any of the 2 switches are showing no drops. It is just the uplink.

It is a closed network with 2 VLANs created. It is justed to replicate data over a iSCSI connection.

Is these packes drops normal? Or what could I do to solve the problem?? HP Support is not telling mutch.

[Matt Hobbs]

Since it is only one that is reporting the drops, the first thing I would do is swap the fibre pairs around on both ends. If the other switch starts reporting the drops instead then it's likely that strand isn't perfect.

Also I'd try swapping the mini-gbics around and then finally the switches. Since you have 2 of everything, you should be able to isolate any faulty hardware.

If it is software/configuraton - are they both running the same version firmware?

100-200 Drops RX a day is quite small, if it isn't causing any noticeable impact I would probably ignore it.

[Ronni Jorgensen]

I have tried to swap the hole switches with 2 new 2650 switches. But I still got drops.
I dont have 2 new fiber cards to swap with. Could it be the fiber patch cables?

The 2 6108 switches are running the same firmware.

Or should I just acceept these 100-200 drops/day?

[Matt Hobbs]

Yes, make sure that the patch cables are also single-mode. I would try swapping them since you saw the same issue with 2650's.

[Paulen]

Hi,

We had a batch with faulty LC drop cables that really were a pain in the @ss. The tip was slightly longer than it should, and the connector didn't click in to place properly. It was very difficult to find the cause of the problems.

Also - ALWAYS clean the tips with a proper cleaning cartridge. It costs almost nothing, and is a very good insurance. We have thousands of patches, and the ones that cause problems are the ones where we used contractors. They don't give a damn.

Always clean - even if the cable is brand new.

Good luck :-)

[Marcus J]

4-5% of drops is considered "normal"

http://www.hp.com/rnd/library/troubleshoot_lan.htm

[Richard Brodie_1]

'4-5% of drops is considered "normal"'

By whom? The link you reference gives 0.02% as a rule of thumb. I personally trigger alerts at half that because it's a round number.

The nominal bit error rate for Gigabit Ethernet is 1 in 10^12, so really one should be aiming for about 0.00001% frame loss.

[Marcus J]

Well i got that figure from HP technical support when i was investigating a similar problem.

Drops is not always an error.

"One key cause of dropped packets is the design of the network topology itself. For example, if twenty 10Mbps clients all try to send data to a 10Mbps server, all connected via a switch, packets can be dropped through no fault of the network device"

[Richard Brodie_1]

OK, I guess it depends what you mean by drops.

Certainly if an intermediate system has to discard packets because of congestion, it's likely a network design, not a hardware problem.

I was specifically referring to 'line noise' type drops, where they are are showing up as CRC or alignment errors on the incoming port. The links themselves should be very reliable: as you say, that doesn't guarantee overall performance.

[Richard Brodie_1]

On reflection, I think that is an important point: if the RX drops aren't genuine receive errors but indicate input buffer overflow (or similar), then no amount of swapping around patch cables is going to make a difference.