caching strategies for high-throughput microservices

our core microservices are getting hammered right now, and the database is definitely feeling it. we're thinking about adding a caching layer, like Redis or maybe Memcached. but I'm fuzzy on the best way to do it. should we use read-through, write-through, or write-behind caching? what's your take on caching for microservices that handle tons of traffic? anything I should look out for