T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A 1 September 2005MC-SSL Simulation 1 Analysis of Scalable Security – MC-SSL Simulation Reducing.

1 September 2005 MC-SSL Simulation1

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Analysis of Scalable Security – MC-SSL Simulation

Reducing excessive cryptographic processing in SSL Connections:

how much can you save?

1 September 2005 MC-SSL Simulation2

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Outline

• Introduction• MC-SSL Background• Methodology• Theoretical Results• Actual Results• Conclusion• Future Work

1 September 2005 MC-SSL Simulation3

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Introduction

• Security processing is CPU intensive• Recent developments on mobile devices

increased its security requirementsex.– Processing stock transaction– Accessing financial institutes

• Hence…the technology development does not fully meet the requires of its applications

1 September 2005 MC-SSL Simulation4

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Introduction(2)

• Similar issues plague battery life of mobile devices in that new applications drain the battery at a faster pace than before– Resolve by scalable features

– Ex. Asus notebooks feature “Asus Power4 Gear Software” that controls CPU speed, LCD brightness, and WLAN

1 September 2005 MC-SSL Simulation5

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

MC-SSL Background

• Developed by James Song – allow third-party (partially trusted) WAP proxy gateway providers– Some mobile devices cannot directly access

data from outside the service provider’s network

– Ex. IP packets need to be transformed into WAP packets before mobile devices are able to view it

1 September 2005 MC-SSL Simulation6

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

MC-SSL Background

VS

MC-SSL:Primary Channel (Confidentiality and Integrity)

MC-SSL: Secondary (Integrity Only) Channel

MC-SSL:Secondary (Clear Text) Channel

SSL Connection:Both Confidentiality and Integrity Protected

1 September 2005 MC-SSL Simulation7

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Methodology

• Java Secure Socket Extension (JSSE) API• Three Elements

– Client– SSL Web Server– Clear Text Web Server

• SSL and Clear Text Web Server on one computer, client on a separate one to avoid interference

1 September 2005 MC-SSL Simulation8

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Methodology – Web Servers

• SSL Web Server Enable Two Cipher Suites– SSL_RSA_WITH_NULL_SHA– TLS_RSA_WITH_AES_128_CBC_SHA

• Clear Text Web Server is an unmodified open-source java Web Server

• Both host MP3 files ranging from 1 to 10 Mbytes, at an interval of 1 Mbyte

1 September 2005 MC-SSL Simulation9

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Methodology – Client

• Initiates connection by enabling one of the two cipher suites offered by the Web Server

• Employs Java Native Interface (JNI) for CPU measurement– C Library

– Collects three measurements• Process’s CPU Time

• Elapsed Time

• CPU Utilization CPU Utilization =

Process CPU Time

-----------------------

Elapsed Time

1 September 2005 MC-SSL Simulation10

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Methodology – Overall

SSL Web Server

Clear Text Web Server

SSL_RSA_WITH_NULL_SHA

Clear Text Channel

TLS_RSA_WITH_AES_128_CBC_SHA

SSL Web Server

Clear Text Web Server

SSL_RSA_WITH_NULL_SHA

Clear Text Channel

TLS_RSA_WITH_AES_128_CBC_SHA

Client

1 September 2005 MC-SSL Simulation11

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Theoretical Results

020

4060

80100

0

20

40

60

80

100

0

10

20

30

40

50

60

70

80

ConfidentialityIntegrity

Savin

gs

1 September 2005 MC-SSL Simulation12

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Theoretical Results

• Based on S. Ravi et al’s “Securing Wireless Data: System Architecture Challenges”

• Assumed linear• Max: 86.5%• Intercept: 30%

3DES 535.9

AES 206.3

SHA 115.4

MD5 33.1

1 September 2005 MC-SSL Simulation13

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Actual Results

1 September 2005 MC-SSL Simulation14

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Actual Results

• Max: 76.4% [vs 86.5%]

• Linear

• Intercept ~35%

• Slope similar, low influence of connection overhead at 10 Mbyte file size

1 September 2005 MC-SSL Simulation15

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Conclusion

• Support the use of scalable secure socket layer connection when CPU capabilities are limited

• Sending large, non-confidential data using integrity only channel can save up to 50% CPU processing power

• Case Study on banking application reveals only 3.4% of data requiring both confidentiality and integrity – 37% CPU saving

1 September 2005 MC-SSL Simulation16

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Conclusion

• Issues– Reintegrating data back together from

separate channels

– Deciding what type of channel for each data

1 September 2005 MC-SSL Simulation17

T H E U N I V E R S I T Y O F B R I T I S H C O L U M B I A

Future Work

• Vary the total file size that is transferred via the network (instead of 10Mbytes)– 8 Mbytes

– 6 Mbytes

– 4 Mbytes, …

• Need to isolate the point which the scheme is ineffective due to overhead

• Experiment on PDA devices (300 MHz, accessing 802.11b/g wireless network)