Elastic Provisioning of Cloud Caches: A Cost-Aware TTL Approach