Managing Complex Systems A Call to Arms and a Plea '''

1
Managing Complex SystemsA Call to Arms! (and a
Plea ...)

• Jacques Sauvé - UFCG
• LANOMS 2007

2
The Problem (1)

• IT-based systems are becoming extremely complex
• Extreme distribution
• Hundreds of thousands of nodes
• Extreme heterogeneity
• Every system is one-off
• Extreme rate of technology change
• Your system isnt client-server? (1980s-1990s)
• Your system isnt Web-based? (1990s-2000s)
• Your system isnt SOA-based? (2000s-?)
• Your arent yet into Web 2.0? (2000s-?)
• Your arent using mashups? (2000s-?)

3
The Problem (2)

• Managing these systems is becoming extremely
  complex
• After decades, we still havent solved basic
  configuration management
• We cant even agree on basic metrics
• Ex. How do you define/measure availability?
• Management maturity evolves (ex. ITSM) but the
  business wants more (governance, lower risk, ...)
• Wow! Youre not using ITIL?
• Wow! Youre not using COBIT?
• Wow! Youre not handling SOX compliance?
• We dont know how to measure risk, calculate
  Return On Investment (ROI), choose SLOs, ...

4
The Problem (3)

• We have a huge system (IT) supporting another
  huge system (a business) and someone comes along
  saying...
• Align one with the other!
• Couple them (oh no!) to give IT visibility into
  the business
• Manage them autonomically ...
• We dont know how to do this ...

5
Why? A Message ...

• We can build complex systems but we dont
  understand them!
• How do users behave?
• How do systems fail?
• What is the emergent behavior of complex systems?

6
About methodology (1)

• We work too much like engineers (build) and not
  enough like scientists (observe)
• Scientists observe nature they let nature talk
  to them
• We should observe our systems to learn
• How do they run?
• How do they break?
• What emergent behavior do they have?
• How does one (IT) affect the other (business)?

7
About methodology (2)

• Using our subjective, finger-in-the-air approach
  wont yield insight
• Insight comes after surprise
• Surprise comes from observation not design
• We should build models later, after we have
  observed
• Roentgen I didnt think, I experimented.
• We dont even agree on metrics yet
• We can't even baseline well
• Because IT services and infrastructure change too
  fast

8
A Call to Arms!

• Let us do more Experimental Computer Science
  using a scientific observation-based methodology
  to gain insight
• Who better than us, the monitoring people?
• This is somewhat of a break from tradition
• Tradition can hinder discovery
• Some can stick to tradition but some must seek
  new ways

9
A Plea

• Make Experimental Computer Science glamorous
• Today, it is (wrongly) ignored
• You have to be a builder to be glamorous
• Add this to the conference Call For Papers
• Lets train our students better
• Lets share data!
• Providers make real data on huge systems
  available
• Must we keep working on toy scenarios?
• What insights will we get?

10
What are the incentives?

• Surely, with time, we will gain new insights and
  make clearer contributions
• Surely, with time, we will come to trust the
  systems we build

11
Thank you.