At the moment of this screenshot 1 or 2 users had idleing sessions in their browsers but nothing happened which explains this heavy workload.
Nevertheless we tested a bit further with instance sizes and came to a fine running solution using
- db.t4g.medium (2 AWS Graviton2 Cores with 4GB RAM, 43-347 MB/s and 2000-11800 IOPS thoughput to harddrive and up to 5 Gibps of network performance) for the postgres-database
- m6a.xlarge (4 AMD EPYC 7R13 Cores with 16GB RAM, 156-1250 MB/s and 6000-40000 IOPS thoughput to harddrive and up to 12,5 Gibps of network performance) for the application server with elasticsearch