Invited Talks
Mitigating the Effects of Internet Timing Faults Across Embedded Network Gateways
Phil Koopman (Carnegie Mellon University, Pittsburgh, PA, USA)
Abstract:Traditional embedded systems such as automobiles and industrial controls are
increasingly being connected to enterprise computing facilities and the
Internet. The usual approach to making such a connection is to install a
gateway node which translates from Internet protocols to embedded field bus
network protocols. Such connections raise obvious security concerns, because
the gateway must guard against attacks on the embedded devices it serves.
For our purposes, we´ll assume that typical enterprise and Internet
vulnerabilities, such as buffer overflows, have already been taken care of.
(Securing devices against traditional attacks is no small matter, but we are
interested in uniquely embedded issues.)
Beyond normal gateway functions, an Internet to embedded gateway must also prevent timing faults and timing attacks from crossing over the gateway to affect the operation of attached embedded systems. An example of timing fault propagation would be severe clumping of messages on the Internet side so that many messages arrive at the gateway all at once, disrupting embedded system operation. While a queue can reduce the loss of incoming data and mitigate network overload, it cannot necessarily protect against timing-related faults on the embedded side of the gateway.
We report simulation results for several mechanisms to mitigate the effects of Internet message timing variations (whether due to faults or malicious attacks) on the performance of networked embedded systems using real-time data. Problems are caused primarily by excessive data delivery delay rather than messages being dropped from arriving clumps. This means that putting a queue in the gateway to manage arriving data clumps is typically worse than using no mitigation mechanism at all. Using a predictive filter seems intuitively better than using a queue, but finding a good generalized predictive filter is also quite difficult.
We believe that managing data streams from the Internet to embedded systems will require careful attention to the nature and time constants of data flowing through the gateway. Moreover, it seems likely that each distinct data stream will need a different set of data management mechanisms and policies at the gateway. In this case, one size does not fit all, making the design of a robust gateway a difficult problem that will require careful modeling of data value behavior for every gateway built.
Beyond normal gateway functions, an Internet to embedded gateway must also prevent timing faults and timing attacks from crossing over the gateway to affect the operation of attached embedded systems. An example of timing fault propagation would be severe clumping of messages on the Internet side so that many messages arrive at the gateway all at once, disrupting embedded system operation. While a queue can reduce the loss of incoming data and mitigate network overload, it cannot necessarily protect against timing-related faults on the embedded side of the gateway.
We report simulation results for several mechanisms to mitigate the effects of Internet message timing variations (whether due to faults or malicious attacks) on the performance of networked embedded systems using real-time data. Problems are caused primarily by excessive data delivery delay rather than messages being dropped from arriving clumps. This means that putting a queue in the gateway to manage arriving data clumps is typically worse than using no mitigation mechanism at all. Using a predictive filter seems intuitively better than using a queue, but finding a good generalized predictive filter is also quite difficult.
We believe that managing data streams from the Internet to embedded systems will require careful attention to the nature and time constants of data flowing through the gateway. Moreover, it seems likely that each distinct data stream will need a different set of data management mechanisms and policies at the gateway. In this case, one size does not fit all, making the design of a robust gateway a difficult problem that will require careful modeling of data value behavior for every gateway built.
Green IT - The Power Saving Challenge and ICT Solutions
Paul J. Kühn (University of Stuttgart)
Abstract:Energy consumption, the finite horizon of conventional fossile energy resources
and maintaining sustainable environmental conditions form the biggest challenges in the near
future. Renewable energy sources like wind, water, solar energy or biomass are limited and
unsteady substitutes and require a radical rethinking of the energy problem. There are two main
solution approaches: power saving and intelligent management of the use of energy.
Both require advanced technologies and a close adaption between energy production
and energy usage. Information Technologies (IT) themselves account for a major energy
consumer by contribution of about 10 % to the global C02 production and will be a target
for power saving but Information and Communication Technology (ICT) are the key for
the intelligent power management.
The first part of the contribution addresses in a systematic way power consumption in ICT on different levels from hardware and device technologies up to application processes, as well as possible approaches and solutions such as new technologies (such as nanotubes), control of power consumption on the chip level, system level and application level by methods of dynamic power supply, adaptive sleep modes, disabling of temporarily unnecessary functionalities, and network virtualization.
In the second part, the purpose and the architectures of energy information networks will be discussed, a comparatively new approach to monitor and to control the consumption of energy depending on the currently available energy sources (such as wind, solar energy or batteries of automotive vehicles), costs for the energy itself and for its transport to the customer. Such energy information networks can be based on existing communication infrastructures (access networks, sensor networks, core networks) which have to be enhanced by other technologies (such as power line communications) and upgraded with respect to security, privacy protection and reliability.
In the final part, the contribution addresses the specific aspect of performance modelling. From this point of view, the issue can be considered as a resource sharing problem. Examples will be given how queuing theory can be used to optimize the use of resources (such as processors, communication links, storage areas, etc.) under stochastic conditions and dynamic scheduling schemes.
The first part of the contribution addresses in a systematic way power consumption in ICT on different levels from hardware and device technologies up to application processes, as well as possible approaches and solutions such as new technologies (such as nanotubes), control of power consumption on the chip level, system level and application level by methods of dynamic power supply, adaptive sleep modes, disabling of temporarily unnecessary functionalities, and network virtualization.
In the second part, the purpose and the architectures of energy information networks will be discussed, a comparatively new approach to monitor and to control the consumption of energy depending on the currently available energy sources (such as wind, solar energy or batteries of automotive vehicles), costs for the energy itself and for its transport to the customer. Such energy information networks can be based on existing communication infrastructures (access networks, sensor networks, core networks) which have to be enhanced by other technologies (such as power line communications) and upgraded with respect to security, privacy protection and reliability.
In the final part, the contribution addresses the specific aspect of performance modelling. From this point of view, the issue can be considered as a resource sharing problem. Examples will be given how queuing theory can be used to optimize the use of resources (such as processors, communication links, storage areas, etc.) under stochastic conditions and dynamic scheduling schemes.
