Speaker
Keith Erickson
(Princeton University)
Description
The National Spherical Torus Experiment Upgrade (NSTXU) at the Princeton Plasma
Physics Laboratory (PPPL) successfully began its first year of operations. NSTXU is a
magnetic fusion device whose major mission is to develop the physics basis for an STbased
Fusion Nuclear Science Facility (FNSF). The STbased FNSF has the promise of achieving the
high neutron fluence needed for reactor component testing with relatively modest tritium
consumption. At the same time, the unique operating regimes of NSTXU can contribute to
several important issues in the physics of burning plasmas to optimize the performance of ITER.
NSTXU uses multiple realtime RedHawk Linux systems based on RedHat Enterprise
Linux 6 (RHEL) for both coil protection and plasma control. NSTXU further uses standard
RHEL6 systems for support services such as housing configuration data and nonrealtime user
interface applications. All of these systems perform critical roles in the success of the NSTXU
project, and it is becoming increasingly apparent that there is a growing risk with respect to
protecting these assets from a security standpoint.
Typically, realtime assets stay hidden behind external protective measures such as
virtual LANs (VLANs) and internal firewalls. With the evolving requirements that organizations
place on all computing assets, these previously sufficient external approaches are no longer
enough to meet all of their goals. Unfortunately, local security policies tend to have an adverse
effect on the deterministic nature of a realtime Linux system, and most policies involve coarse
and inflexible settings.
As part of an ongoing initiative to protect computing assets from both malicious and
accidental threats, NSTXU developed multiple approaches to blend tight controls with careful
study of realtime effects. Included here will be coverage of how NSTXU managed to balance
the primary purpose of the Linux systems with additional security constraints, including using
Security Enhanced Linux (SELinux), specific firewall settings, Linux “capabilities” (that is,
specific superuser privileges that do not require superuser access), and numerous other
security measures. In all cases where a security change negatively affected realtime
performance, that change was either mitigated or reverted. What remains is a grouping of safe
alternatives that show that both security and realtime determinism are both practical and useful.
Author
Keith Erickson
(Princeton University)