<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Complex Systems on VinhMDev</title><link>https://vinhmdev.com/topics/complex-systems/</link><description>Recent content in Complex Systems on VinhMDev</description><generator>Hugo</generator><language>en</language><lastBuildDate>Thu, 02 Jul 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://vinhmdev.com/topics/complex-systems/index.xml" rel="self" type="application/rss+xml"/><item><title>Paper 06: Phase Transitions: Why Overloaded Systems Freeze Instead of Slowing Down</title><link>https://vinhmdev.com/posts/paper-06-phase-transitions-why-overloaded-systems-freeze-instead-of-slowing-down/</link><pubDate>Thu, 02 Jul 2026 00:00:00 +0000</pubDate><guid>https://vinhmdev.com/posts/paper-06-phase-transitions-why-overloaded-systems-freeze-instead-of-slowing-down/</guid><description>&lt;p&gt;






 
 
&lt;figure&gt;&lt;img
 src="https://static.vinhmdev.com/posts/1-series-the-illusion-of-control/2-phase/6-post/The-Goodput-Cliff.jpg"
 alt="A statistical-mechanics analysis of overload: why goodput drops off a cliff instead of degrading linearly, why a jammed system delivers less than it did right before the jam, and which early-warning signals surface in latency telemetry before the collapse."
 class="mx-auto my-0 rounded-md" loading="lazy" decoding="async"
 /&gt;
&lt;/figure&gt;
&lt;/p&gt;
&lt;h2 id="i-abstract" class="relative group"&gt;I. Abstract &lt;span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100"&gt;&lt;a class="group-hover:text-primary-300 dark:group-hover:text-neutral-700" style="text-decoration-line: none !important;" href="#i-abstract" aria-label="Anchor"&gt;#&lt;/a&gt;&lt;/span&gt;&lt;/h2&gt;&lt;blockquote&gt;
&lt;p&gt;Linear intuition predicts that 10% more load costs roughly 10% more latency. Landau&amp;rsquo;s classification of phase transitions, Kleinrock&amp;rsquo;s queueing divergence, and the physics of traffic flow predict something else: a loaded system crosses a load-driven transition with three measurable signatures. First, a goodput cliff — useful output does not degrade, it drops discontinuously. Second, capacity drop — the jammed state delivers less than the system did right before jamming, and that state sustains itself. Third, critical slowing down — the variance and lag-1 autocorrelation of latency rise well before the mean does, which makes the collapse observable in advance. The engineering consequences: define sustainable capacity at the knee of the latency curve rather than at the benchmark peak, aim monitoring at second-order statistics rather than means, and when the system is already jammed, shed load far below nominal capacity instead of trimming it by degrees.&lt;/p&gt;</description></item><item><title>Paper 05: Self-Organized Criticality: Why Systems Drift Toward the Edge of Collapse</title><link>https://vinhmdev.com/posts/paper-05-self-organized-criticality-why-systems-drift-toward-the-edge-of-collapse/</link><pubDate>Wed, 01 Jul 2026 00:00:00 +0000</pubDate><guid>https://vinhmdev.com/posts/paper-05-self-organized-criticality-why-systems-drift-toward-the-edge-of-collapse/</guid><description>&lt;p&gt;






 
 
&lt;figure&gt;
 
 








 
 &lt;picture
 class="mx-auto my-0 rounded-md"
 
 &gt;
 
 
 
 
 &lt;source
 
 srcset="https://vinhmdev.com/The-Cost-Optimized-Criticality_9022592046581105705_hu_6e0f054b8da3602a.webp 330w,https://vinhmdev.com/The-Cost-Optimized-Criticality_9022592046581105705_hu_5c9466cbd57da48c.webp 660w
 
 ,https://vinhmdev.com/The-Cost-Optimized-Criticality_9022592046581105705_hu_fa8102b79b361372.webp 1024w
 
 
 ,https://vinhmdev.com/The-Cost-Optimized-Criticality_9022592046581105705_hu_513c76074cb4ba0c.webp 1320w
 "
 
 sizes="100vw"
 type="image/webp"
 /&gt;
 
 &lt;img
 width="1376"
 height="768"
 class="mx-auto my-0 rounded-md"
 alt="A mathematical analysis of why cost-optimized distributed systems drift toward a critical state, where incident sizes follow a power law and a minor trigger can set off an avalanche that spans the cluster."
 loading="lazy" decoding="async"
 
 src="https://vinhmdev.com/The-Cost-Optimized-Criticality_9022592046581105705_hu_c51093d9df04d418.jpg" srcset="https://vinhmdev.com/The-Cost-Optimized-Criticality_9022592046581105705_hu_ccad25efe1f03927.jpg 330w,https://vinhmdev.com/The-Cost-Optimized-Criticality_9022592046581105705_hu_c51093d9df04d418.jpg 660w
 
 ,https://vinhmdev.com/The-Cost-Optimized-Criticality_9022592046581105705_hu_7ecafd5cb7df2325.jpg 1024w
 
 
 ,https://vinhmdev.com/The-Cost-Optimized-Criticality_9022592046581105705_hu_a74fd3f4bbee14fc.jpg 1320w
 "
 sizes="100vw"
 
 /&gt;
 &lt;/picture&gt;
 


&lt;/figure&gt;
&lt;/p&gt;
&lt;h2 id="i-abstract" class="relative group"&gt;I. Abstract &lt;span class="absolute top-0 w-6 transition-opacity opacity-0 -start-6 not-prose group-hover:opacity-100"&gt;&lt;a class="group-hover:text-primary-300 dark:group-hover:text-neutral-700" style="text-decoration-line: none !important;" href="#i-abstract" aria-label="Anchor"&gt;#&lt;/a&gt;&lt;/span&gt;&lt;/h2&gt;&lt;blockquote&gt;
&lt;p&gt;The dashboard reads 80% CPU, 0% errors. Linear intuition concludes: 20% headroom remains, the system is safe. The Bak-Tang-Wiesenfeld sandpile model and Highly Optimized Tolerance theory conclude otherwise: a distributed system does not sit still inside its safety region — it drifts toward the critical point, and cost-optimization pressure is the engine driving it there. At that point, incident sizes follow a power law: the same minor trigger that is harmless today spreads into a cluster-wide avalanche tomorrow. The paradox: the system sits at the edge not because some law of nature forces it, but because you — sprint by sprint of cost cutting and rising utilization — pushed it there. This paper presents the mathematics of that drift, and how to hold a system back from the edge.&lt;/p&gt;</description></item></channel></rss>