Cloudoscopy: services discovery and topology mapping

We define and study cloudoscopy, i.e., exposing sensitive information about the location of (victim) cloud services and/or about the internal organisation of the cloud network, in spite of location-hiding efforts by cloud providers. A typical cloudoscopy attack is composed of a number of steps: first expose the internal IP address of a victim instance, then measure its hop-count distance from adversarial cloud instances, and finally test to find a specific instance which is close enough to the victim (e.g., co-resident) to allow (denial of service or side-channel) attacks. We refer to the three steps/modules involved in such cloudoscopy attack by the terms IP address deanonymisation, hop-count measuring, and co-residence testing. We present specific methods for these three cloudoscopy modules, and report on results of our experimental validation on popular cloud platform providers. Our techniques can be used for attacking (victim) servers, as well as for benign goals, e.g., optimisation of instances placement and communication, or comparing clouds and validating cloud-provider placement guarantees.