In case the IRF channel breaks and both switches decide to become active, IRF works according to a similar algorithm with Cisco VSS. One of the switches retains its role of the main switch, the second switch goes to the recovery state (Recovery-state). In this state, it disables all ports except IRF and those ports for which it is manually specified that they do not need to be disabled. As soon as the IRF channel is restored, the switch that was in the recovery state will reboot. After the reboot it will become a slave.
The schemes of Multi-active detection are also very similar to the ones we discussed in Cisco VSS. HPE IRF supports the following options:
- LACP MAD (similar to Cisco Enhanced PAgP)
- BFP MAD (similar to Cisco IP BFP)
- ARP MAD (Gratuitous ARP containing the active device ID is used)
- ND MAD (using NS Neighbor Discovery packets within the IPv6 framework)
HPE recommends the use of LACP or BFP MAD, as these are the fastest mechanisms. ARP and ND MAD are slower and require the use of STP (which is a bit unexpected). By the way, the mechanisms use different logic to select the switch that will remain the main one (wizard), so HPE does not recommend that they be used together (namely, LACP with the rest).
Now let's see how the HPE IRF differs from Cisco VSS.
First, IRF technology is supported on a wider range of switches. In fact, this is the main stacking technology, which is available on the relatively cheap A3100 and expensive 12900 modular switches. Only one series of switches can be stacked, though with a small exception (5800 and 5820 series, as well as 5900 and 5920 series will work together).
Secondly, IRF technology allows stacking up to nine switches (for some models this value is limited to four). Two switch topologies are possible within IRF: bus and ring.
A variant with a ring connection is recommended because it is more fault-tolerant. If one connection is broken, we will not get a situation with two active IRF switch groups.
Judging by the description of the technology, after the switches have joined together on the IRF stack, they start exchanging some hello-packages to build a common stack topology. Then, based on this topology, packets are transferred within the IRF stack. Unfortunately, no more detailed information could be found.
Before concluding the IRF review, it is necessary to mention the development of this technology - enhanced IRF (eIRF). eIRF allows you to build a more hierarchical structure, including two levels - kernel and access level (Clos architecture is used).
Switches at the kernel level (the Spine level for the Clos architecture) will perform management functions (such switches are called Controlling Bridges - CB). And in fact, the IRF stack runs on them. Switches at the access layer (the Leaf level for the Clos architecture) will only provide port expansion and traffic only (these switches are called Port Extenders (PE)). PE switches (in some documents these switches are referred to as PEX) do not have local switching functions and absolutely all packets will be transmitted to the kernel switches. At the moment there can be two kernel switches and up to 30 access level switches. All these switches will look like one big device. Does it remind you of anything? Of course, a similar solution for Cisco is Instant Access.
An example of IRF technology implementation
Before we get to the final part, I would like to point out once again that some of the questions for me were left out of the picture, especially the issue of packet transmission within the IRF stack.
Conclusion
I would like to give a summary table of the technologies considered earlier. But I would like to point out that I will not determine "who is the coolest in this sandbox". First of all, both technologies are proprietary and in fact their use is a consequence of the choice of this or that vendor as a manufacturer of solutions for the construction of network infrastructure. Secondly, each item can be discussed for a long time. For example, IRF is supported on a wider range of models. And Cisco has its own stacking technologies on the younger models. VSS supports only two switches. And why more? And so on. Third, I have considered only the general functional aspects, leaving such important issues as stability, ease of maintenance, troubleshooting, etc. out of the picture.
- Cisco VSS HPE IRF
- Where 4500X, 4500E, 6500E, 6800 3100, 3600, 5120 etc. are supported
- Number of devices that can be combined 2 9
- Switching with state saved Yes (SSO/NSF) Yes
- Switching speed in case of failure of the main switch 200-400 ms 50 msec