vendredi 27 février 2015

The Black Swan

I finally got the opportunity to read the best-selling book The Black Swan by the Lebanese-American professor Nicholas Nassim Taleb. I am not a big fan of best-sellers, but this book is unique in its kind. Indeed, the author develops his thoughts across several disciplines (risk analysis, finance & economy, statistics, epistemology, history, philosophy...) in a funny and entertaining writing style.

Although I found the book a bit long and redundant, the exposed idea is very interesting. Black Swan symbolizes rare events that have big impact, such as tsunami, financial crisis, music hit or world war. I will talk about the main idea that interested me in the book: our blindness to Black Swans.

We believe that we made reliable predictions using verified models, until we get surprised by a sudden collapse of Wall Street... Taleb explains that our blindness to Black Swans is due to cognitive biases. These biases are strongly rooted in our biology because the evolution of humans have taken place during thousands of years in a linear environment with no significant deviations. Today, things have changed with free markets, globalization and technology. Our world is not linear anymore! Indeed, one can become rapidly millionaire by investing in a lucky startup or can cause a financial crisis by a trading transaction...

The first bias is the confirmation error or the induction problem. We tend to generalize from empirical observations thus building fragile theories that are very vulnerable to improbable events. Europeans always thought that all swans were white, since all the swans they observed were white, until the day that they discovered a black swan in Australia. Serial killers have a clean record, until the day they murder their victims!
Moreover, generalizing from empirical observations get worse when we mis-estimate the representational value of the observed sample. Consider this conclusion: "In my startup I will adopt the X management style, because CEOs of successful startups recommend it in management magazines."  Unfortunately, this is a Black-Swan-exposed conclusion, because CEOs of bankrupted startups are not invited to write about this management style in business magazines!

The second bias is the narrative fallacy. We constantly need to put events in a reasonable story with causal correlations in order to mentally digest them thus making us blind to random events escaping this pattern. This explains why sometimes we fail to predict events in he future, but typically find logical interpretation to them after their occurrence. Everybody can explain the causes of the second world war, but can we predict the next war? how many people would have imagined the "great" war a week before its start?

The third bias is emotional, and can have different aspects. It strongly reminded me of Treblinka a novel by J.F. Steiner which is based on the true story of an extermination camp in Poland. Steiner argues that Jews of Poland heard horrible witnesses of many escaped prisoners from camps, but however unconsciously refused to believe the existence of such a cruelty and thus were lead easily to death camps without resistance. they emotionally got blinded by a fake hope. "Human nature is not programmed for Black Swans."

The last bias is "tunneling", also the expert syndrome. In fact, experts that make prediction models do not see risks outside their expertise domain, their tunnel. A financial company would hire the best mathematicians to find prediction models in order to reduce risks of market fluctuations, but forget to put in place a Recovery Plan for its Information System/Datacenter thus completely risking its existence. We live in the era of butterfly effects, where a terrorist from Afghanistan can bring down two sky scrapers in the USA, leading to important changes in cloud hosting business in Europe. We should look to what's happening outside of our tunnel.

When you read with Taleb about the flaws of predictions and the impact of randomness in history, you might fall rapidly in obsessional skepticism and non-decision attitude. Well, that is not the author's intention. He is inviting us to understand the limits of our knowledge in this non-platonic world and act upon it, to be conscious of our UnKnowledge and stop epistemological arrogance in order to avoid negative Black Swans. Furthermore, Taleb explains how to get more exposed to positive Black Swans and benefit from them!

PS: This book was dedicated to Benoit Mandelbrot, the famous mathematician who gave more importance to randomness and chaos in scientific approach with his fractal theory.

mardi 24 février 2015

Hybrid networks strategy

The uses on corporate WAN networks are continually evolving with the rise of collaborative tools (VoIP, Visio, IM...), public cloud based applications, connected objects, personal usages... Indeed, according to a recent study, companies are witnessing the following:
  • 20% is the yearly global growth of traffic 
  • 30% of traffic is critical
  • 40% of traffic's destination is Internet
These evolutions are driving companies to rethink their WAN strategies. Hybrid networks emerge as an alternative to a full provider based MPLS network. With sites connected simultaneously to both MPLS and Internet, hybrid strategy is about taking the best from the two worlds. In fact, for each application the best link is chosen according to the following criteria: 

Below is an example of a Hybrid strategy where critical private applications and real time traffic are routed on MPLS, non critical private applications are offloaded on Internet and public cloud based applications as well as surf are routed directly to Internet. Note that security scheme must be carefully adapted: on site firewalls, cloud based proxies... Note also that network availability is naturally increased with a hybrid architecture.

Routing traffic can be network level based using PBR or application level based using more intelligent technologies such as Ipanema, Stonesoft, Fortinet, Riverbed... The advantage of such technologies, depending on vendors, is the possibility to integrate security functions such as firewall, compression functions and application QoS functions with traffic monitoring tools.

Hybrid strategy can be even pushed further with multiple Internet and MPLS connections with different providers. 

Finally, in order to build a better hybrid strategy, some questions must be asked:
  • Is my security scheme adapted to my hybrid architecture? how can I have the same level of security? a homogeneous security policy?
  • Can I, or my service provider, manage the potential complexity of hybrid networks?
  • Are my managed solutions with my provider compatible with hybrid architecture? what are the technical limitations?
  • When calculating all the costs of hybrid networks, is it financially interesting for my case?