Mature cloud Wi-Fi offerings have gone through few phases already. They started with bare-bones device configuration from the cloud console and over the years matured into meaty management plane for complete Wi-Fi access, security and complementary services in the cloud.
Alongside these phases of evolution, optimizing the cost of operation of the cloud backend has always been important consideration. It is critical for cloud operators and Managed Service Providers (MSPs). This cost dictates what end users pay for cloud Wi-Fi services and whether attractive pricing models (like AirTight's Opex-only model) can be viable in the long run. It is also important to the bottom line of the cloud operator/MSP.
Posed with the cost question, one would impulsively say that cost is driven by the capacity in terms of number of APs that can be managed from a staple of compute resource in the cloud. That is an important cost contributor, but not the only one!
What do the cost models from cloud operation reveal?
We have monitored cloud backend operation costs for past several years. Based on that data, we have built some cost models. These models have led to the discovery of factors that are significant cost contributors. Identifying the cost component is a major step towards reducing it. The cost reduction is often implemented by the combination of technology and process innovations.
Draining the cost out of cloud
This one is no brainer for anyone with head in the cloud. Scalability generally refers to number of APs that can be managed with a unit of compute resource. Higher scalability helps reduce the cost. Enough said.
As the customers of diverse scales (10 APs to 10,000 APs) are deployed in the cloud and at diverse paces, it often results into unused capacity holes in the provisioned compute resources. The capacity holes are undesirable, because the cloud operator or MSP has to pay for them, but they don't get utilized towards managing end user devices.
The unused capacity problem needs to be solved at two points in time: Initial provisioning and re-provisioning. Clearly, when new customers are deployed, you try to fit them in the right sized capacity buckets. Assuming they love your product, they will then deploy more and start to outgrow their capacity buckets (but you also cannot over-provision, else there will be capacity hole from the beginning). This is the re-provisioning time. At that time, the cloud architecture and processes need to be able to seamlessly migrate customers to bigger capacity buckets.
The very reason customers have chosen to go with cloud is because they want plug-n-play experience. As such, the patience level of the cloud customer is often lower than the one choosing the onsite deployment option. This necessitates higher level of plug-n-play experience to avoid support calls.
There are various points in the life cycle that have high tendency to generate support calls. One point in time is when devices connect to the cloud, or let’s say, not able to connect to cloud. Another critical time is during software upgrades. The issues also often arise during re-provisioning as discussed above when customers are migrated between compute resources. The cost of attending to support calls can be a significant factor if these experiences are not super smooth. Additional complexities arise when APs are sold through channel, but cloud is operated by vendor or another MSP.
The pricing logic behind reducing personnel cost at MSP is as follows. The end user is eliminating the onsite personnel cost by migrating to cloud, and hence paying less on TCO basis. When the experience is not smooth, this cost is transferred to the personnel at the cloud operator or MSP. The cloud operator and MSP cannot make money if they pick up significant part of this cost on their head.
Certain features such as high availability and disaster recovery have potential to give rise to latent resources. Latent resources are different from capacity holes discussed before. Latent resources are like insurance in that they don't get utilized most of the time, but they need to be maintained in great shape. Brute force implementation of these redundancy features has been found to be significant cost contributor to cloud operation.
For any cloud services platform, the above pain points are exposed after years of operational experience and teething pain with diverse customer deployments. That is why, it would be appropriate to say that there are two parts to the viable cloud operation - one is the computing technology that enables complete management features and the other is operational maturity. You overlook any one of them and the cloud can become unviable for operator/MSP and customers in the long term.