I question the architecture and approach of this much caching. Because this is memory, I assume it's for a read-only cache? That said: most read heavy services

I question the architecture and approach of this much caching. Because this is memory, I assume it's for a read-only cache? That said: most read heavy services only need at most 10% of their working set cached, in memory. The hardest problem in caching is figuring out what that set is and how to control consistency. Not how to store it. So, having this much cache seems to imply that you think you're going to have large amounts of read intensive data to cache. Or else you intend to cache your entire working set? Either way, you'll have a single point of failure (either in a single server, single datacenter, or single geographic area). Any large operation can tell you that the problem becomes 90% network and stuff like a CDN become far more important than how fast and big your cache is. Perhaps a nice technical writeup on your architecture would silence the pundit inside me?