dong hyuk woo georgia tech hsien-hsin “sean” leegeorgia tech

23
Analyzing Performance Vulnerability due to Resource Denial-Of-Service Attack on Chip Multiprocessors Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” Lee Georgia Tech

Upload: kerry-oliver

Post on 31-Dec-2015

27 views

Category:

Documents


0 download

DESCRIPTION

Analyzing Performance Vulnerability due to Resource Denial-Of-Service Attack on Chip Multiprocessors. Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech. Cores are hungry. “Yeah, I’m still hungry..”. Cores are hungry. More bus bandwidth? Power.. Manufacturing cost.. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

Analyzing Performance Vulnerabilitydue to Resource Denial-Of-Service

Attackon Chip Multiprocessors

Dong Hyuk Woo Georgia Tech

Hsien-Hsin “Sean” Lee Georgia Tech

Page 2: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

2

Cores are hungry..

“Yeah, I’m still hungry..”

Page 3: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

3

Cores are hungry..

• More bus bandwidth?– Power..– Manufacturing cost..– Routing complexity..– Signal integrity..– Pin counts..

• More cache space?– Access latency..– Fixed power budget..– Fixed area budget..

Page 4: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

4

Competition is intensive..

“Mommy, I’m also hungry!”

Page 5: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

5

What if one core is malicious?

“Stay away from my food..”

Page 6: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

6

Type 1: Attack BSB Bandwidth!

• Generate L1 D$ misses as frequently as possible!– Constantly load data with a stride size of 64B

(line size)– Memory footprint: 2 x (L1 D$ size)

Normal CoreNormal Core

L1 I$L1 I$ L1 D$L1 D$

Malicious CoreMalicious Core

L1 I$L1 I$ L1 D$L1 D$

Shared L2$Shared L2$

Page 7: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

7

Type 2: Attack the L2 Cache!

• Generate L1 D$ misses as frequently as possible!

• And occupy entire L2$ space!– Constantly load data with a stride size of 64B

(line size)– Memory footprint: (L2$ size)

• Note that this attack also saturates BSB bandwidth!

Page 8: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

8

Type 3: Attack FSB Bandwidth!

• Generate L2$ misses as frequently as possible!

• And occupy entire L2$ space!– Constantly load data with a stride size of 64B

(line size)– Memory footprint: 2 x (L2$ size)

• Note that this attack is also expected to– saturate BSB bandwidth!– occupy large space of the L2 cache!

Page 9: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

9

Type 4: LRU/Inclusion Property Attack• Variant of the attack against the L2 cache• LRU

– A common replacement algorithm

• Inclusion property– Preferred for efficient coherent protocol

implementation

• Normal core accesses shared resources more frequently.

set

way

Page 10: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

10

To be more aggressive..

• Class II– Attacks using Locked Atomic Operation

• Bus locking operations– To implement Read-Modify-Write instruction

• Class III– Distributed Denial-of-Service Attack

• What would happen if the number of malicious threads increases?

Page 11: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

11

Simulation

• SESC simulator• SPEC2006 benchmark

Number of Cores 4

Issue width 3

L1 I$2-way set associative 32KB cache with 64B line (1 cycle hit latency)

L1 D$2-way set associative 32KB cache with 64B line (1 cycle hit latency)8-entry MSHR

BSB data bus B/W 64 GBps (2GHz * 256 bits)

L2$8-way set associative 2MB cache with 64B line (14 cycle hit latency)Shared MSHR

FSB bandwidth 16 GBps

DRAM latency 100 ns

Page 12: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

12

Vulnerability due to DoS Attack

Normal Normal

vs.

Page 13: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

13

Vulnerability due to DoS Attack

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Norm

alize

d I

PC

astar lbm mcf soplex harmonic mean

Load/BSB Load/L2 Load/Incl. Load/FSB Atomic/BSB Atomic/L2 Atomic/Incl.

High L1 miss rate

High L2 miss rate

Page 14: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

14

Vulnerability due to DDoS Attack

Normal Normal

vs.

Normal Normal

Page 15: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

15

Vulnerability due to DDoS Attack

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Norm

alize

d I

PC

Load/BSB Load/L2 Load/Incl. Load/FSB Atomic/BSB Atomic/L2 Atomic/Incl.

1 malicious thread 2 malicious threads 3 malicious threads

Page 16: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

16

Suggested Solutions

• OS level solution– Policy based eviction– Isolating voracious applications by process

scheduling

• Adaptive hardware solution– Dynamic Miss Status Handler Register (DMSHR)– Dedicated management core in many-core era

Page 17: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

17

DMSHR

Entry 0Entry 1Entry 2Entry 3Entry 4Entry 5Entry 6Entry 7

MSHR full

Compare

Counter

MSHR full

Decision from monitoring

functionality

Page 18: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

18

Conclusion and Future Work

• Shared resources in CMPs are vulnerable to (Distributed) Denial-of-Service Attacks.– Performance degradation up to 91%

• DoS vulnerability in future many-core architecture will be more interesting.– Embedded ring architecture

• Distributed arbitration

– Network-on-Chip• A large number of buffers are used in cores and

routers.

Page 19: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

19

Q&A

Grad students are also hungry..

Please feed them well..Otherwise, you might face Denial-of-??? soon..

Page 20: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

Thank you.

http://arch.ece.gatech.edu

Page 21: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

21

Difference from fairness work

• They are only interested in the capacity issue

• They might be even more vulnerable..– Partitioning based on

• IPC• Miss rates

– They may result in a guarantee of a large space to the malicious thread.

Page 22: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

22

Difference between CMPs and SMPs

• Degree of sharing– More frequent access to shared resources in CMPs

• Sensitivity of shared resources– DRAM (shared resource of SMPs) >> L2$ (that of

CMPs)

• Different eviction policies– OS managed eviction vs. hardware managed

eviction

Page 23: Dong Hyuk Woo Georgia Tech Hsien-Hsin “Sean” LeeGeorgia Tech

23

Difference between CMPs and SMTs

• An SMT is more tightly-coupled shared architecture.– More vulnerable to the attack

• Grunwald and Ghiasi, MICRO-35– Malicious execution unit occupation– Flushing the pipeline– Flushing the trace cache

– Lower-level shared resources are ignored.