Posts on Technology

The complexity of feature-rich systems

Posted by Diego Assencio on 2015.10.05 under Technology (General)

Anyone who works with technology knows intuitively that as more features are added to a system, its complexity increases rather fast and often in unpredictable ways. Given that the more complex a system is, the more susceptible it is to incorrect or suboptimal functioning due to an improper interaction of its features, one is lead to wonder: how fast does the complexity of a system actually grow with the addition of new features?

Even if an exact quantification of the complexity of a given system could be properly defined, its computation would certainly require us to understand the system in its entirety, which is not exactly what we are aiming at here. Consider instead this simple question: for a system having $n$ features, how many pairwise feature interactions are possible?

The question may sound too abstract, so let us motivate it first. For that, consider a car. If one steps on the gas pedal while the parking brake is activated, the car may even move, but slower than if the parking brake would be released. The car may even get physically damaged as a result of operating in this mode for too long. This means two features of the system (the gas pedal and the parking brake) lead to a suboptimal functioning of the car when in the state "gas pedal pressed down" with "parking brake activated".

The example just given illustrates just one of the many possible interactions which can happen between the features of a car. A car is a complex system, and understanding how its features interact is necessary for its proper operation. If a car has $n$ features in total, the number of pairwise feature interactions $N^{(2)}$ is the total number of possible combinations of two from the $n$ features: $$\displaystyle N^{(2)} = \left(\matrix{n \\ 2}\right) = \frac{n!}{2!(n-2)!} = \frac{n(n-1)}{2} = \frac{1}{2}(n^2 - n)$$ This result is already alarming: the number of pairwise feature interactions grows quadratically with the number of features. The natural question which follows is: what is the number of possible interactions $N^{(3)}$ involving three features? The answer is the total number of possible combinations of three from the $n$ features: $$\displaystyle N^{(3)} = \left(\matrix{n \\ 3}\right) = \frac{n!}{3!(n-3)!} = \frac{n(n-1)(n-2)}{6} = \frac{1}{6}(n^3 -3n^2 + 2n)$$ Things are now even worse: the complexity of our system grows cubically with the number of features. Let us then go ahead and compute the total number $N$ of possible interactions over any possible subset of features of the system: $$\begin{eqnarray} N &=& N^{(2)} + N^{(3)} + \ldots + N^{(n)} \nonumber\\[5pt] &=& \displaystyle \left(\matrix{n \\ 2}\right) + \left(\matrix{n \\ 3}\right) + \ldots + \left(\matrix{n \\ n}\right) \nonumber\\[5pt] &=& \sum_{k=0}^n \left(\matrix{n \\ k}\right) - \left(\matrix{n \\ 1}\right) - \left(\matrix{n \\ 0}\right) \label{post_bd2985d293fcd5959b3b4f4f36d3965a_eq_deriv_N} \end{eqnarray}$$ We can now use the binomial theorem to compute the sum on the last line of the equation above: $$\displaystyle (x + y)^n = \sum_{k=0}^n \left(\matrix{n \\ k}\right)x^{n-k}y^k$$ Taking $x = y = 1$, we have then that: $$\displaystyle (1 + 1)^n = 2^n = \sum_{k=0}^n \left(\matrix{n \\ k}\right) \label{post_bd2985d293fcd5959b3b4f4f36d3965a_eq_binom_theorem}$$ The right-hand side of equation \eqref{post_bd2985d293fcd5959b3b4f4f36d3965a_eq_binom_theorem} is exactly the sum which appears on the last line of equation \eqref{post_bd2985d293fcd5959b3b4f4f36d3965a_eq_deriv_N}. Therefore: $$\boxed{N = 2^n - n - 1}$$ As $n$ grows, the value of $N$ quickly becomes dominated by the term $2^n$, i.e., the number of possible feature interactions grows exponentially with the number of features.

Although simplistic, the analysis above has dramatic consequences to the development and testing of new technologies. As more features are added to a system, the number of possible interactions between its features grows so fast that it becomes quickly impossible for the human mind to fully comprehend the system as a whole. Testing its usage under all possible scenarios may become not just difficult, but impractical. As a result, unforeseen problems may end up appearing in the production version of the system — some of which may be really embarrassing for those who designed them.

Dealing with iptables errors on a Raspberry Pi

Posted by Diego Assencio on 2015.04.25 under Technology (Raspberry Pi)

If you use iptables on your Raspberry Pi but after a firmware upgrade start getting an error similar to the one below:

iptables v1.4.14: can't initialize iptables table nat': Table does not exist (do you need to insmod?)


you can handle the problem by downgrading the firmware to a version which does not have this issue. Doing this is really simple. Just go to the rpi-update repository page and choose a commit which was created before the day in which you upgraded the firmware of your Raspberry Pi. Click on this commit and then copy the sequence of characters which appears right below the commit message (for those who are unfamiliar with git, this weird-looking sequence of characters is called a commit hash and uniquely identifies a commit). To be more explicit, all you have to do is look for something like this:

commit ba4304f7bec24d5f0a4150f09a37884240f8926d2


Now open a terminal on your Raspberry Pi and run:

sudo rpi-update <commit-hash>


Here is an example with the commit hash shown above:

sudo rpi-update ba4304f7bec24d5f0a4150f09a37884240f8926d2
`

Reboot the Raspberry Pi and see if iptables is now working. In the unlikely case that the problem persists, just choose another commit and redo the steps above.

How much power does a Raspberry Pi consume?

Posted by Diego Assencio on 2014.01.04 under Technology (Raspberry Pi)

I have recently purchased a USB current and voltage tester and decided to check how much power my Raspberry Pi typically consumes. The tester device works simultaneously as a voltmeter and as an ammeter: it goes in series with the USB device and has an LED display which shows the current and voltage being used (see figure 1).

The Raspberry Pi website claims the model B uses between 700-1000mA (depending on the connected peripherals) and needs 5V to operate. I have measured, however, significantly smaller values for the current even when the Raspberry Pi was under heavy load.

 Fig. 1: Measuring the power consumption of the Raspberry Pi. The pictures show examples of the measured current ($I = 0.36\textrm{A}$) and voltage ($V = 4.96\textrm{V}$) respectively.

With the measured current $I$ and the voltage $V$, one can compute the power consumption using: $$P = VI$$ For $V = 5\textrm{V}$ and $I = 700\textrm{mA} = 0.7\textrm{A}$, the power consumption would then be: $$P = 5\textrm{V} \times 0.7\textrm{A} = 3.5\textrm{W}$$ However, the highest current I registered was $I_{\max} = 0.43\textrm{A}$ and the highest voltage was $V_{\max} = 4.95\textrm{V}$ (these maximum values were seen during boot time and under heavy load). The maximum power consumption I saw was then: $$P_{\max} = V_{\max} I_{\max} \approx 2.13\textrm{W}$$ which is significantly less than the $3.5\textrm{W}$ computed above. Under normal conditions, my Raspberry Pi consumed between $1.8\textrm{W}$ and $2.0\textrm{W}$.