190 likes | 197 Views
Explore information leakage in third-party compute clouds, placement and co-residence determination, load measurement, keystroke timing attack, and novel elements in cloud computing security.
E N D
Lecture 5: Cloud Security: what’s new? Xiaowei Yang (Duke University)
Recap • Exploring information leakage in third-party compute clouds • Placement • Determining co-residence • Inferrence
Placement • Launching test instances • Determining the correlation between instance placement and IP addresses • Launching many probe instances in the same availability zone
Determining co-residence • Traceroute
Cross-VM information leakage • Load measurement: Prime-Trigger-Probe • B: buffer of size b; s: cache line size • Prime: Read B at s-offset • Trigger: busy-loop until swapped out • Probe: measure the time it takes to read B again at s-offset • If it takes long • If it does not take long
Load-based co-residence detection • Send http requests to a target VM • Do load measurement • High • Low
Estimating traffic rates • High traffic rates high load
Keystroke timing attack • Hypothesis • On an idle machine, • High load spike keystroke input • Timing between high load spikes timing between keystrokes • Timing between keystrokes infers password
Summary • Co-residence information leak • Defending against it is hard
Overview • New threats • New research opportunities
New threats • A more reliable alternative to botnets • If cloud computing is cheaper and more reliable than botnets, use cloud • Brute-forcer • Resource sharing and interference • Placement, inferrence • Reputation fate sharing • Spammers block other legitimate services • An FBI raid
Novel elements • Protecting data and software is not enough Activity pattern needs protection as well • Reputation attribution • A longer trust chain • Competitiveness business may co-locate
Is mutual auditability a solution? • Provider audits customer’s activities • Customer audits what a provider provides • enables attribution of blame
New opportunities • Cloud providers should offer a choice of security primitives • Granularity of virtualizations • Physical machines, LANS, clouds, or datacenters • Mutual auditability • Provider audits customer’s activities • Customer audits what a provider provides • enables attribution of blame • Studying cloud security vulnerabilities
Next • Discovering VM dependencies using CPU utilization • Question to ponder: can this technique be used a security attack?
Interesting techniques • Inference technique • Auto-regressive modeling: use past samples to predict future values • Compute distances of AR models • Models with similar coefficients are closer • K-mean clustering • Perturbation to improve inference accuracy
Security attacks • Achieving co-residence • Do load measurements • Figure out service correlations • DoS all related services