Collaborative Research: Design and Analysis of
High-Performance, Energy-Efficient, and Secure Clusters
Supported by NSF CCF-0541384
Last Updated: January 31, 2007
Summary
Clusters have emerged as the most effective solution
to design high performance servers, which are
increasingly being deployed in supporting a wide
variety of Web-based services.
Along with high and predictable performance,
optimization of energy consumption in these servers
has become a serious concern due to their high power
budgets.
In addition, the critical nature of many Internet-based
services mandate that these systems should be robust
to attacks from the Internet, since numerous security
loopholes of cluster servers come to the forefront.
Therefore, design and analysis of high performance,
energy-efficient and secure clusters is crucial for
the next-generation cluster systems not only from
academic and industrial standpoints but from socio-economic
and environmental standpoints.
Although some initial investigation on cluster energy
consumption and security has appeared recently,
an in-depth design and analysis of a cluster
interconnect considering the three parameters
above have not been undertaken.
On the other hand, such an investigation is extremely
challenging because there are numerous controllable
factors across many dimensions and frequently these
factors are conflicting.
The proposed research attempts to address three closed issues
in a ground-up fashion starting from the basic cluster components
to the entire system.
First, design and analysis of a complete cluster interconnect or
a system area network (SAN) will be conducted.
The work will encompass the study of the recently proposed
InfiniBand Architecture (IBA) framework.
The interconnect fabric includes the switches, Network Interface
Cards (NICs) or Host Channel Adaptors (HCAs), and various
network topologies.
Second, we will develop a framework to provide enhanced
security in terms of confidentiality, integrity, and
availability in clusters, considering all sources of
vulnerabilities including operating system/software,
communication hardware, user-level communication and
network protocols.
We will also propose a novel authentication mechanism
with marginal performance overhead and no InfiniBand
packet format change.
Finally, we will investigate the adaptation of IBA
in a cluster server and propose efficient memory
management techniques and energy conservation schemes
in a whole cluster system including disk systems,
SANs and individual cluster nodes.
A comprehensive simulation testbed will be developed to
evaluate these design concepts along with real experiments.
We will provide hierarchical analysis of cluster systems,
thus enabling assessment of the effects of individual
components such as disk subsystems and a cluster network.
We are planning to design all components to be plugged
into the simulation testbed to assess their impacts
on performance, energy consumption, and security of
the entire cluster system.
Publications
-
Y. Jin, E. J. Kim, and K. H. Yum,
"A Domain-Specific On-Chip Network Design for Large Scale Cache
Systems," to be presented in IEEE HPCA-13.
Ki Hwan Yum