Secure Cloud Computing: The Monitoring Perspective

Size: px

Start display at page:

Download "Secure Cloud Computing: The Monitoring Perspective"

Brianne Armstrong
8 years ago
Views:

1 Secure Cloud Computing: The Monitoring Perspective Peng Liu Penn State University 1

2 Cloud Computing is Less about Computer Design More about Use of Computing (UoC) CPU, OS, VMM, PL, Parallel computing Multi-tenant, Resource consolidation, Computing as a service, IT management, Cloud Computing = UoC innovations + 2

3 Security issues due to UoC innovations Isolation and inference channels Trust-minimizing computing Accountability 3

4 Isolation and inference channels Physical isolation disables resource consolidation Logical isolation = physical sharing Logical isolation leads to inference channels - Explicit data flows - Implicit information flows - Covert channels 4

5 Trust-minimizing computing Tenants apps do not need to trust OS. Tenants VMs no need to trust provider s VMMs/hardware. Tenants data no need to trust apps. 5

6 Accountability Make data accountable Make information flows accountable Make code and control flows accountable Make SLAs accountable 6

7 Security monitoring is essential Without monitoring, accountability cannot be achieved. Monitoring plays a critical role in inference control. What to monitor? - Data flows - Control flows - Information flows - Data invariants - Cross the isolation boundaries 7

8 State of the art Coarse-grained monitoring is mature and widely deployed. Fine-grained monitoring is not very practical. - Dynamic taint analysis is still offline (3x-100x) - Inlined monitoring is expensive - 8

9 Why so hard to make fine-grained monitoring practical? The Collapse of Moore's Law is a fundamental reason. CPU core s speed simply cannot maintain its rapid exponential rise using standard silicon technology. So unless we rewrite an app to do parallel computing, the app s response time will not decrease in future. So inlined monitoring will still be a pain in future. 9

10 Non-Blocking Concurrent Security Monitoring -- Let monitor code run on other cores during idle time 10

11 Motivation Program execution Security checking Sync. Program execution Security checking Core 1 Core 2 Inlined checking Concurrent checking 11

12 Problem 1: App Heap Buffer Overflow Monitoring 12

13 Straightforward (but inefficient) attempts Canary-based checking [Cowan and Pu 1998] User threads Canary addresses Monitor thread Attempt1: Lock-based red-black tree Monitoring blocks program execution Attempt2: Lock-free hash table [Shalev & Shavit 2006]. Complex operations and Contention 13

14 Cruiser Architecture Custom lock-free data structures and non-blocking algorithms to collect canary addresses. 14

15 Technical hurdle Theorems on impossibility of lock-free non-blocking synchronization. Please refer to our PLDI 11 paper. 15

16 Performance SPEC CPU2006 5% with Eager Cruiser, 12.5% with Lazy Cruiser 5, 000 whole-heap checks per second 16

17 Scalability Apache Negligible average overhead Cruising cycle < 80 us (12, 500 times/second) 17

18 Problem 2: Kernel Heap Buffer Overflow Monitoring 18

19 Out-of-the-VM Architecture Core 1 Core 2 Core 3 19

20 Hybrid VM monitoring Architecture Kernel address space Guest VM1 Secure address space Guest VM2 Network drivers File systems Hooks Entry code Exit code Entry code Exit code Heap metadata Monitor VMM The cruising cycle = 7ms 20

21 Technical hurdles 1. Race conditions 2. Self-protection Please refer to our NDSS 12 paper. 21

22 Performance Overhead SPEC CPU06 Execution time perlbench bzip2 gcc mcf gobmk hmmer sjeng libquantum h264ref omnetpp SIM-Kruiser Less than 3% Normalized to the execution time of original Linux astar xalancbmk Kruiser geo. mean 22

23 Scalability - Apache Requests per second Original SIM-Kruiser Kruiser Concurrency Throughput for varying numbers of concurrent requests. 23

24 Final remark Exciting innovations on concurrent monitoring are yet to come. - Data flows - Control flows - Information flows - Data invariants - Cross the isolation boundaries 24

25 Thank you! Acknowledgment: The works mentioned in this talk are supported by NSF, AFOSR MURI, and ARO MURI. 25

Achieving QoS in Server Virtualization

Achieving QoS in Server Virtualization Intel Platform Shared Resource Monitoring/Control in Xen Chao Peng (chao.p.peng@intel.com) 1 Increasing QoS demand in Server Virtualization Data center & Cloud infrastructure