TESTING DUPLEX MISMATCH ON CML

[JUANNN]

Hello,

I am running CML 2.7.2 and using IOSvL2 switches with the following topology:

Is a very simple lab, and my goal is to see the Collision and Late Collision counters increase when I send data on both ways at the same time. 

The 2 switches are already configured, and the interfaces are UP/UP and one in HALF DUPLEX (SW2) and the other one in FULL DUPLEX (SW1). As expected, I am receiving the CDP message of DUPLEX MISMATCH, but I can still forward data (as it should be). However, since the Half-Duplex interface should be seeing collisions that in reality aren't occuring, it should be dropping some data packets (because it will stop transmitting and then continue to transmit again as a result of seeing collisions). 

But whenever I show the interface information on the Half Duplex side, the Collision Counters are still in 0!! I tried every packet size and over 1000 packets at one time... (pinging simultaneously from both ways) and nothing!! 

Is rare because I tried to mismatch the MTU of the interfaces and I was able to see the GIANTS counter going up, but for some reason on DUPLEX MISMATCH I can't see the Collision Counter growing. Any one knows what is going on?? Thanks,

Juan

[MHM Cisco World]

try ping repeat 10000 times from each side and check count 

MHM

[balaji.bandi]

I am sure you wont be achieved 100% results, since this is virtual - so you can 100% simulate like hardware Layer1 and Layer 2 here.

[JUANNN]

But CML uses the Cisco IOS images that I use on hardware, and like I said is rare that I am able to see GIANTS for example, but not Collisions counters

[Ramblin Tech]

@balaji.bandiis right, CML does not provide 100% faithful emulation of L1-signaling over L0-media to the IOS/XE/XR/NXOS images it hosts. Not sure what you mean by "CML uses the Cisco IOS images that I use on hardware", but the CML node images definitely are not the same as any image that runs directly on an ISR or Catalyst switch/router.

IOS images are built to run on specific "platforms". That is, an IOS image is built by compiling and linking Platform Independent (PI) software components to Platform Dependent (PD) components, with PD components being tailored specifically to the underlying platform by the Development Engineers (DEs). This is why you cannot install and run a Cat8000 image on a Cat9000 platform: the hardware platforms are different and so the platform dependencies are all wrong (eg, SDK calls, APIs, mappings of hardware components to memory locations, etc). When Cisco's IOS Release Operations group builds images with the same IOS version numbers across multiple platforms, all of the platforms' various dependencies are accounted for by the automated build process and you end up with releases with similar PI capabilities running on many different platforms, but those images are not interchangeable with each other.

In the case of IOSv/IOSvL2, they are their own platforms, as platforms do not have to be physical h/w. That is, DEs designed a Virtual Machine (VM) as the platform with certain h/w capabilities emulated (RAM, disk, Ethernet NICs, console, etc). When an IOSv/IOSvL2 image is built, IOS PI code is compiled and linked to PD code that was written to interface to the emulated h/w of the VM. In yet another layer of emulation with IOSvL2, an emulated NIC then transmits Ethernet frames over an emulated L1/L0 Ethernet network.

There are no actual Ethernet networks underlying CML's connectivity between nodes, only Linux Bridge constructs to facilitate "transmitting" frames between nodes via inter-process communication channels. The IOSvL2 platform wraps an emulation layer around the Linux Bridge to present an Ethernet-like behavior to the PD code. In half-duplex scenarios, it attempts to emulate L1 signaling associated with CSMA-CD operation where there is no actual CSMA-CD taking place. This L1/L0 emulation is good, but it is not 100% perfect.

[JUANNN]

In half-duplex scenarios, it attempts to emulate L1 signaling associated with CSMA-CD operation where there is no actual CSMA-CD taking place. 

That makes sense, thanks.  

[MHM Cisco World]

Did you ping 10000 from both sides ??

MHM

[JUANNN]

I did, no luck.