There is more to distributed database system design than CAP, however. For example, low latency (speed) is a chief concern for many architects. If you read the Amazon Dynamo[66] paper, you’ll notice a lot of talk about availability but also Amazon’s latency requirements. For a certain class of applications, even a small latency change can translate to a large costs. Yahoo’s PNUTS database famously gives up both availability on normal operation and consistency on partitions in order to squeeze a lower latency out of its design.[67] It’s important to consider CAP when dealing with distributed databases, but it’s equally important to be aware that distributed database theory does not stop there.