On Thu, 17 Nov 2011 08:18:39 +0000 Seb Rupik snr@ecs.soton.ac.uk wrote:
Hi Morten, I'm still very much in the learning phase of SNMP so I couldn't really make any suggestions on any models. I do however have four 3750G-48TS switches which I'd happily wire together for the purpose of testing any relevant patches.
Thanks, Seb, that could come in handy; however, I currently have several ideas that could help solve the apparently 3750-related timeout issues, while I have no good idea how to model the various stacking paradigms that are in the wild.
If you feel that 3750 stack members that don't appear as modules in NAV are more than just a cosmetic problem, I would really like some suggestions or a discussion on how to model things, before we blindly make some patch.
Warning: Brain dump follows.
NAV used to have a terrible data model for HP virtual stacks, which we entirely ripped out of the code at one point. An HP virtual stack is basically just a bunch of autonomous switches which take up only a single IP address for management - one still needs to communicate with each stack member individually through the stack commander as a proxy, by using a modified community for each member.
My best suggestion for modeling those (though we never implemented support for those again) was automatically adding the stack members as separate IP devices sharing the stack commander's IP, but with modified communities. That wouldn't necessitate changing NAV's code all over the place, which was necessary under the old model. This still left the problem of presentation, as network operators still wanted to display all the ports of a stack on a single NAV page.
I guess Cisco clustering works much the same way as HP virtual stacking, and could be modeled the same way.
Although I have no experience with it, stacking of 3750's and Cisco VSS sound like they're basically the same thing model-wise: Via SNMP, a stack and a VSS will just look like a big switch with a lot of ports.
The 3750 stack may be simple to model. I don't know the hardware very well, but I'm guessing the 3750's aren't modular, so the switch ports in each will appear to be part of the chassis. Thus, the 3750 stack members could probably be modeled as modules in NAV.
But a VSS could consist of several modular switches. I'm guessing NAV would find the slot modules of each stack member and say that they're all modules in the same big box, but NAV wouldn't know about the multiple chassis, and has no good way of modeling that two-level hierarchy.
At the moment, the NAV model basically says that "a Netbox/IP Device = A chassis" and "a module = some replaceable sub-unit inside a chassis". Should we add an extra chassis-level, and say that the netbox/IP device is just a model of the details needed to communicate with some network host?