This section explain two approaches: invalidating from web servers, and invalidating from storage servers.
First, this section did not explain how they work. A diagram for each will help.
Secondly, the article mentions two problems of invalidating from web servers. I do not understand why the two problems only apply to invalidating from web servers. For example: The first problem: “The packet overhead is larger due to it being a broadcast”. The problem also exists for the second approach: invalidating from storage servers, right?