Let's take a look at the practice of deploying ThingsBoard on AWS. It describes the cheapest deployment option as well as the most flexible and expandable. **The calculation is based only on the cost of the instances and database, not including the load balancer, TB license fee, etc.** This article will deal with thingsboard with a hybrid database type. The deployment options (prices are per month): * 1 instance with type m6a.xlarge (4 CPUs and 16G RAM) - the most basic version we recommend. ==**Its cost is approximately $145 per month.**== All services are deployed on the server. * **Pros:** lowest cost. For a low-load system - **10000 data points/sec**. * **Cons:** Low scalability and availability. Postgres is backed up using a script (this may cause additional CPU usage). Potential for data loss. Not enough resources for Cassandra backup services. * 1 instance with type m6a.xlarge (4 CPUs and 16G RAM - $145) and RDS db.t4g.medium (2 CPUs and 4G RAM - $52). This version has the same server and Postgres deployment in RDS. ==**Its cost is approximately $197 per month.**== * **Pros:** it is still cheap. Great for low-load systems - **10000 data points/sec**. Allows high-level Postgres database administration (usage metrics, automatic backups). The resources that Postgres would use can be used to back up Cassandra. But these resources may not be enough. It all depends on the size of Cassandra. * **Cons:** Low scalability and availability. Potential for Cassandra data loss. * 1 instance with type m6a.xlarge (4 CPUs and 16G RAM - $145) + 3 instances with type m6a.large (2 CPUs and 8G RAM - $73 per 1) and RDS db.t4g.medium (2 CPUs and 4G RAM - $52). This version has the same server and Postgres deployment in RDS and 3 servers for the Cassandra database. ==**Its cost is approximately $416 per month.**== * **Pros:** Great for low and medium load systems - **10000-20000 data points/sec**. Allows high-level Postgres database administration (usage metrics, automatic backups). The Cassandra cluster ensures the high availability of data even if one of the nodes fails. More resources for the thingsboard and third-party services on the main instance will receive a greater load. * **Cons:** Low scalability. * 3-node m6a.xlarge (4 CPUs and 16G RAM - 145$ per 1) for Kubernetes cluster + EKS ($70) and RDS db.m6g.large(2 CPUs and 8 RAM - $134) ==**Its cost is approximately $639 per month.**== * **Pros:** Great for medium-load systems with **20000 data points/sec**. The cluster has high scalability and availability. Microservice architecture. * **Cons:** more costly than previous options. * 3-node m6a.2xlarge (8 CPUs and 32G RAM - $291 per 1) for Kubernetes cluster + EKS ($70) + RDS db.m6g.large (2 CPUs and 8G RAM - $134) + AWS MSK Kafka ($148) + Redis (cache.t4g.small $80) ==**Its cost is approximately $1305 per month.**== * **Pros:** Great for high-load systems with **30000 data points/sec**. * **Cons:** most costly option. The "**data points/sec**" load is an approximate indicator to be guided by. Much also depends on rule chains, the number of scripts in rule chains, the size of messages, and the number of devices. Therefore, deployment options are chosen individually.