A bit of a puzzling blog post. F5's CMP implementation has been around a long time, and all the issues noted rise from some of the fundamental design decisions. TMM instances are self contained and don't communicate with their peers. So the overhead of sharing any information like what would be required for applying real time connection limits is high, forcing either period aggregation at a reasonable frequency (the active health check issue), or by not sharing it at all (the connection limits issue). So it's not like a new version or updated hardware is going to make this behavior go away.