| Organodynamics | Grant Holland, Apr 25, 2014 |
| Slide: What problem am I addressing? | |
| Context: - My (recent) job: Principal Systems Engineer -
My
Employer: Sun Microsystems, Inc. -
Product:
Unix Servers and Server Farms Challenge: -
How do you re-configure 1000+ servers that must appear to operate as one computer system to the users? -
In real-time, continuously? -
Task has grown beyond the grasp of human actors (Unix admins and systems engineers) | What do I mean by Òre-configure? -
Continuously reorganize the
configuration of the components -
Shared resource management -
Dynamic
job/thread balancing -
Dynamic
reallocation of shared resources -
Parallel
threads and processes -
Predictive
resource allocation Two principle concerns: 1. Continuous system reorganization 2. Unpredictability
of the required system reorganization over time |
| In other words, How do you do automated,
dynamic data center management? -
AKA, Òcloud infrastructure managementÓ in todayÕs lexicon -
In short, Sun needed to develop an Òoperating systemÓ that performed continuous, automated, real-time system organization over a distributed collection of N servers, for any N>0. | How
do we classify this class of problem? -
This
is a problem in system
organization -
Over
time -
That
is, dynamical
system organization -
More
accurately, a problem in dynamical
systems reorganization Research revealed that the only existing class of systems that exhibited this kind of self-organization is biological systems. |
Notes: