Steelhead Appliance

[joshtseng]

Hi Michael,

My advice is pretty obvious--keep the Steelhead LAN interface set at auto/auto.

I do understand what you're talking about, and as I as I stated earlier, I myself have experienced the exact same thing you have experienced.  All I can do is offer an explanation.

The Ethernet interface is implemented in hardware, and the behavior you are experiencing is a result of the hardware manufacturer's interpretation and implementation of the IEEE spec.  In the case of the Steelhead's in-path Ethernet interface, that manufacturer is not Riverbed, but one of Riverbed's component suppliers, or one of that component supplier's upstream suppliers.  I'm pretty sure the same is true for the Cisco switch.

The IEEE spec states to use half duplex if there is any difficulty negotiating duplex, and it seems for some implementations that happens if there is any unexpected behavior from the peer interface, even when not in autonegotiate mode, and when hard-set to full duplex.  So it seems one of the interfaces is seeing something that it doesn't like from the other interface.  I don't know if it's the Riverbed or Cisco interface, but one of them is still ending up at half duplex, even when explicitly told to use full duplex.

Josh

[michaelthb]

hi Josh

Thanks for all your help so far.

 This scenario is slightly different. We are setting the Cisco 2950 Fast ethernet port to 100/F and then power up the Steelhead with the LAN port set to Auto and monitoring the result of the auto negotiation.

What we experience is that the LAN port negotiates to 100/F with no errors. We then set the LAN port to 100/F, and then experience errors. 

This is very strange as under both conditions, we apprear to be running at 100/F but the auto negotiation method is the only one to not show errors.

Your advice will be much appreciated.

Michael 

[joshtseng]

Hi Michael,

I've personally had the same exact experience that you've encountered.  I have no explanation other than that for some Ethernet manufacturers, setting to 100/F doesn't necessarily mean 100/F; they still will go to 100/H if they see behavior they don't like from the other interface.

Josh

[michaelthb]

Hi Josh

Thanks for your reply.

 I hear what you are saying about the autonegotiation standards. However, the autnegotiation process appears to be fine in this scenario. They negotiate to 100/F with no errors. What I'm concerned about is that when I fix the LAN port on the Riverbed to what it autonegotiates to, errors start incrementing. 

[joshtseng]

Michael,

You can blame the IEEE for Ethernet autonegotiation standards that are broken and ambiguous.

This has been a huge frustration encountered in many deployments.  The problem is that Ethernet manufacturers have implemented the autonegotiation process differently in response to the ambiguous standards.  You just have to find a combination that works for your specific set of equipment.

Josh

[michaelthb]

Hi,

I've done a Riverbed implementation at a customer. While checking the ethernet, I noticed that the errors on the Riverbed LAN port were incrementing. The Riverbed was connected to a Cisco Catalyst 2950. The port on the Switch was fixed to 100/F.

I then changed the port to Auto/Auto on the Riverbed and the errors stoped incremented. And the LAN port on the Riverbed negotiated to 100/F to the switch port.

The LAN port on the Riverbed negotiated to 100/F on auto and ethernet was clear. But when I fixed the port on the Riverbed to 100/F, errors increment quickly.

Can anyone explain what could cause this to happen?