"Although computer memory is no longer expensive, there's always a finite size buffer somewhere. When a big piece of news arrives, everybody sends a message to everybody else, and the buffer fills"
About this Quote
Mandelbrot points to a stubborn truth about complex systems: abundance in one component does not erase bottlenecks elsewhere. Even when memory is cheap and plentiful, queues, channels, and processors still have finite capacity. When an unusually large signal hits the system and people broadcast to everyone at once, traffic fans out explosively and overwhelms whatever buffers stand in its path.
The image is both technical and social. Computers drop packets and back up queues; users experience delays, errors, and outages. Markets do something similar. A surge of orders after major news strains the order book, a buffer of resting bids and offers, and liquidity vanishes just when demand peaks. In both cases the average load looks manageable, but the rare, clustered surges dominate the risk. That is a hallmark of the heavy-tailed, bursty behavior Mandelbrot emphasized in his critiques of Gaussian assumptions in finance and engineering. Systems tuned for the mean fail when the tails wag the dog.
There is also a fractal sensibility at work. Shocks propagate as branching trees: A tells B and C, who each tell two more, and soon a local ripple becomes a network-wide cascade. Self-similar bursts appear at many scales, saturating small buffers first and then larger ones. Cutting the price of RAM does not change the shape of these cascades, because contention moves to CPU interrupt queues, network interfaces, disk I/O, or human attention itself. The buffer simply relocates.
The practical lesson is design and governance under humility. Expect surges rather than smoothing them away with comforting averages. Use rate limits, backpressure, sharding, circuit breakers, and graceful degradation to keep failures contained. Model loads with power-law tails, not only normal noise. And remember the social analogue: attention and trust are finite buffers too. When everybody messages everybody, misinformation and panic can overflow them just as surely as a network queue.
The image is both technical and social. Computers drop packets and back up queues; users experience delays, errors, and outages. Markets do something similar. A surge of orders after major news strains the order book, a buffer of resting bids and offers, and liquidity vanishes just when demand peaks. In both cases the average load looks manageable, but the rare, clustered surges dominate the risk. That is a hallmark of the heavy-tailed, bursty behavior Mandelbrot emphasized in his critiques of Gaussian assumptions in finance and engineering. Systems tuned for the mean fail when the tails wag the dog.
There is also a fractal sensibility at work. Shocks propagate as branching trees: A tells B and C, who each tell two more, and soon a local ripple becomes a network-wide cascade. Self-similar bursts appear at many scales, saturating small buffers first and then larger ones. Cutting the price of RAM does not change the shape of these cascades, because contention moves to CPU interrupt queues, network interfaces, disk I/O, or human attention itself. The buffer simply relocates.
The practical lesson is design and governance under humility. Expect surges rather than smoothing them away with comforting averages. Use rate limits, backpressure, sharding, circuit breakers, and graceful degradation to keep failures contained. Model loads with power-law tails, not only normal noise. And remember the social analogue: attention and trust are finite buffers too. When everybody messages everybody, misinformation and panic can overflow them just as surely as a network queue.
Quote Details
| Topic | Internet |
|---|
More Quotes by Benoit
Add to List
