240 likes | 254 Views
Self-organizing systems & methods for managing Banking IT Infrastructure. Jim Harlow 5-24-2005. About NetFuel, Inc. Founded in 1999 Founders - Jim Harlow & Bjorn Ahlblad Investors Bank of America, Inc. Citadel Investments, LLC Borland
E N D
Self-organizing systems & methods for managing Banking IT Infrastructure Jim Harlow 5-24-2005
About NetFuel, Inc. • Founded in 1999 • Founders - Jim Harlow & Bjorn Ahlblad • Investors • Bank of America, Inc. • Citadel Investments, LLC • Borland • Mission – self-healing banking systems (networks & datacenters). • Goal – deliver an overlay “Guardian Operating System” for banking systems.
The problem, and how we defined it Business problem – regardless of ALL circumstances, revenue generating transactions, in order of priority, must never be lost, unnecessarily delayed, or neglected. • A system of computing elements is a continuum of non-linear multi-dimensional datasets. Hyper-dimensional manifolds that need to be simplified for human consumption. • Dimensions include orthogonal elements - state, state-change-vector, time, location, line of business and priority. • By looking for aberrant systemic behavior patterns, that vary from expected patterns in “real-time”, we can return the system to nominal operation by adjust the aberrant before it arrives in a Byzantine or fault-stop state.
High level constraints Global Controller • Autonomous • Programmable • Scalable • Auditable • Deterministic • Fault tolerant
Personnel constraints • 98% of command center staff are male • Statistically higher concentration of them are • Red-green Color-blind • Mildly Dyslexic • Most are not schooled in computer science. • Their compensation plans are generally based uponn their efficiency as “repeaters and dispatchers”.
We needed a few things in order to build a commercial product • A granular model to work from • Neuropharmacology and self-learning algorithms. • A lexicon to describe it • Autonomic Computing lexicon via IBM • A means of visualizing it • Kohonen Self Organizing Maps (SOM) • Line of Business “Dashboard”
Audit of commercial attempts • Bayesian methods • Wavelets • Stochastic modeling • Stateless reporting • Stateful textual alerting The permutations are nearly endless… We hired people from “GMD Fokus” in Berlin, Peakstone in Silicon Valley, and a number of other experts
NetFuel implementation • Multi-agent system using six (6) different types of software-based agents • Federated Directory Service • Federating Message Bus • Atomic work-flow • Kohonen Self-learning algorithms • Kohonen SOM
What’s special about our agents • In-line updating of methods, policies, and constraints. • On-platform or off-platform. • Request new methods and policies if constraints hinder goals. • Intelligent Static Agents – transitory broad scope • Autonomous Intelligent Static Agents – stand-alone and self-guiding. • Mobile Agents – migratory narrow scope • Mobile Intelligent Agents - migratory broad scope • Autonomous Mobile Intelligent Agents – self guiding
Agent Mobility and Redundancy • Agents are mobile. They can signal the ARE to serialize their state and move them to another ARE. • A set of “server” agents can cooperate to manage a set of objects. The transaction integrity provided by the directory service is leveraged to bid on ownership of objects and for bidding should one server fail or need to move. • This allows agents to reorganize to spread load or to work around network problems.
Federating Message Bus The message bus supports proxy connections. With a proxy connection a 3rd party ARE is used to facilitate connections between two AREs. In DMZ and Internet environments, proxy connections allow security administrators to only allow one or two designated AREs to connect to AREs on the internal network. All other AREs must then connect through the designated AREs.
Kohonen Self-organizing Maps • Representations of a vector-based space illustrating rates of changes within optimal functional domains. • Distributed systems have more than single influences upon their behavior – sometimes in excess of 20 dimensions to optimal functional realms. • SOM represents hyper-dimensional manifolds in simple two-dimensional spaces. • Successfully used in OCR and computerized medical diagnosis systems equipment already in use for over a decade.
CPUs Load Balancing Load Balancing Web HTTP Servers Web HTTP Servers Authentication & Authorization Trading Authentication & Authorization Trading
Disk I/O Load Balancing Web HTTP Servers Authentication & Authorization Trading
“P” Threads Load Balancing Web HTTP Servers Authentication & Authorization Trading
File handles Load Balancing Web HTTP Servers Authentication & Authorization Trading
JVM Load Balancing Web HTTP Servers Authentication & Authorization Trading
App Srvr Load Balancing Web HTTP Servers Authentication & Authorization Trading
EJBs Load Balancing Web HTTP Servers Authentication & Authorization Trading
Pkt I/O Load Balancing Web HTTP Servers Authentication & Authorization Trading
Load Balancing Load Balancing Load Balancing Load Balancing Load Balancing Load Balancing Load Balancing Web HTTP Servers Web HTTP Servers Web HTTP Servers Web HTTP Servers Web HTTP Servers Web HTTP Servers Web HTTP Servers Authentication & Authorization Authentication & Authorization Authentication & Authorization Authentication & Authorization Trading Trading Trading Trading Trading Composite
Zero-in on State of devices over time where policy has altered behaviors over many dimensions simultaneously Load Balancing Web HTTP Servers Authentication & Authorization Trading
CCE+ (Global Modeler) Agent Store Policy Store Agent Authoring tool NetFuel Message Bus Config Mgr + PKI Signer Policy Authoring tool Policy Tester Network Simulator Global Modeler, CCE- State Manager, Kohonen SOM Policy Publishing x
CCE- (Regional Modeler) Agent Store Policy Store NetFuel Message Bus Agent12 Agent3 Agent2 Agent10 Agent9 Agent5 Agent6 Agent1 Agent7 Agent8 Agent4 Modeler/ARE State Manager x New Agent