Cisco recently announced that they are releasing CML-P, which is version two of the product formerly known as VIRL. First of all, I’ve seen the product demoed and helped with feedback on it, it looks stunning! The architecture looks great, it’s fully leveraging APIs and it’s an entirely different beast than VIRL. This is a great product and I want to see it succeed. Unfortunately, this product is never going to be as successful as it could be. Why?
CML-P, where P stands for Private, supports a maximum of 20 nodes. This is supposed to be a differentiator to the the -E version, which is for enterprises that wish to run this product at larger scale, including support. First of all, I don’t agree that a node limit is the proper way to differentiate -P from -E. That can be done through support, training and other means.
CML-P’s competition is going to be GNS3 and EVE-NG. These are freely available, but also offer paid versions with a more advanced feature set. There is no node limit with these products. You can run as much as your server can handle. If CML-P is going to compete with these products, it will be at disadvantage, since these products aren’t locked to a specific number of nodes. There’s no doubt in my mind that people are willing to pay for CML-P, but the node limit makes it a lot less attractive.
Why Is a Node Limit Bad?
I believe a product like this should support you through all certification levels that Cisco offers, meaning up to CCIE level. For those of you that have taken a CCIE lab, you know that the topologies can be quite large. Why are the topologies so large? It’s about being able to fit many technologies into a topology without having too many protocols/features configured on each device. When you want to learn a technology, you don’t want 10 other protocols configured, as when you’re troubleshooting, you won’t know what’s broken and you may end up with faulty logic, because of something else that was configured. There are vendors that offer very large troubleshooting topologies that candidates use to hone their troubleshooting skills. To summarize, a simulation tool should be able to take you through your CCIE studies.
Another reason it’s bad is that if you want to learn a new technology, let’s say SD-WAN, you want to build a fairly realistic scenario. That means that you will probably end up with one or two datacenters and something like three sites. A datacenter often has several modules such as Internet Edge, WAN Edge, Core, Services and Distribution/Access layer. For the DCs alone, you’ll probably be close to 20 nodes. Then you have your Branch sites where you may want to have local switches and so on. This means that you probably end up with 30-40 nodes for a full scale simulation.
Another reason why the node count adds up, especially for something like the SD-WAN scenario I just mentioned, is that you may want to simulate a service provider. That is, you want to simulate Internet and MPLS and possibly add packet loss etc. This means you need to add even more nodes.
But Daniel, do we really need this many nodes? Who has a server to support that many nodes? There are plenty of people that have decent servers to support up to a 100 nodes or more, depending on what image they run.
Everything about CML-P looks really awesome. Except for the name perhaps. I believe this product could easily overpower its competitors due to its architecture and automation capabilities. However, it will always be held back by the artificial node limit. I look forward to seeing this product in action and I sincerely hope the team behind CML-P will reconsider and remove the node limit, or at least push it to something more sensible like 50 devices. What do you think?