Paper Review. Greyness Failure: The Achilles' Heel Of Cloud-Scale Systems

MAD questions 1. Is this related to robust-yet-fragile concept?
This notion of masked latent faults afterwards causing large disruptions reminds live of the robust-yet-fragile systems concept. Robust-yet-fragile is well-nigh highly optimized tolerance. If yous optimize your tolerance exclusively for crash failures only non partial/gray failures, yous volition live really disappointment when yous are faced alongside this unanticipated mistake type.

A skillful lawsuit hither is the glass. Glass (think of car spectacles or gorilla glass) is truly really tough/robust material. You tin sack throw pebbles, in addition to fifty-fifty bigger rocks at it, in addition to it won't intermission or scratch, well, upwards to a signal that is. The drinking glass is really robust to the anticipated faults (stressor) upwards to a point. But, transcend that point, in addition to and then the drinking glass is inward shambles.  That shows an unanticipated stressor (a dark swan lawsuit inward Taleb's jargon) for the glass: a ninja stone. The ninja stone is basically a slice of ceramic that yous accept from the spark plug, in addition to is denser than glass. So if yous gently throw this really petty slice of ceramic to your auto window, it breaks inward shambles.  

This is called a robust-yet-fragile structure, in addition to this is truly why nosotros had the Titanic disaster. Titanic, the ship, had really robust panels, only over again upto a point. When Titanic exceeded that signal a petty chip (with the iceberg hitting it), the panels broke into shambles, really much similar the drinking glass coming together ninja stone. Modern ships after Titanic, went for resilient, instead of robust (yet fragile) panels. The resilient panels curvature easier, only they don't intermission equally miserably. They yet concord together to the confront of an extreme stressor. Think of plastic; it is less robust only to a greater extent than resilient than glass. 

The robust-yet-fragile upshot is likewise known equally highly optimized tolerance. If yous optimize tolerance for i anticipated stressor, yous teach out really vulnerable to to a greater extent than or less other unanticipated fault. (Much similar the unopen Australian ecosystem.)

2. Is fuzzy logic applicable here?
It seems similar instead of binary detectors which output good for yous or failed, it is ameliorate to lead maintain detectors that laissez passer on probabilities in addition to confidence to their decisions. So I idea the phi accrual detectors should live a relevant operate to consider for detecting gray failures. I don't know if at that topographic point are whatever other fuzzy detectors operate for identifying latent failures.

Update: Ryan Huang, i of the authors of the work, left a comment alongside insightful answer to my questions. In the response, he includes links to followup operate equally well. https://docs.google.com/document/d/18Du33J1v3wOhqj-Vcuv5-wPnaweGhvnWFzenmHoxVcc/edit

Share this post