Cache Invalidation Strategies: Which one for a highly distributed system?

Hey everyone, I'm really struggling with cache invalidation in a system that's all over the map, geographically speaking. We've got multiple services, databases, and caches spread out everywhere. Right now, we just use a time limit for cache entries, but it feels pretty crude and sometimes we end up with old data. I've been checking out pub/sub methods and write-through/write-behind techniques to get better consistency. What's worked well for you guys when it comes to keeping caches consistent in these complicated, distributed setups? Any total nightmares or unexpected wins with particular strategies? I'm trying to figure out how to keep things consistent without killing performance.