<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Serving-Systems on Jaehun's Blog</title><link>https://jaehun.me/categories/serving-systems/</link><description>Recent content in Serving-Systems on Jaehun's Blog</description><generator>Hugo</generator><language>ko-kr</language><lastBuildDate>Fri, 11 Sep 2026 09:31:03 +0900</lastBuildDate><atom:link href="https://jaehun.me/categories/serving-systems/index.xml" rel="self" type="application/rss+xml"/><item><title>Deadline-Aware Adaptive Prefill Chunking for Efficient Large Language Model Serving</title><link>https://jaehun.me/posts/deadline-aware-adaptive-prefill-chunking-for-efficient-large-language-model-serving/</link><pubDate>Fri, 11 Sep 2026 00:00:00 +0900</pubDate><guid>https://jaehun.me/posts/deadline-aware-adaptive-prefill-chunking-for-efficient-large-language-model-serving/</guid><description>&lt;p&gt;&lt;a&#10; href="https://arxiv.org/abs/2609.07883v1"target="_blank"&#10; class="inline-flex items-center gap-1"&#10; &gt;논문 링크&lt;svg class="h-3 w-3 flex-shrink-0" id="external-link" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"&gt;&lt;path fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M15 3h6v6m-11 5L21 3m-3 10v6a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2V8a2 2 0 0 1 2-2h6"/&gt;&lt;/svg&gt;&#10; &lt;/a&gt;&lt;/p&gt;&#10;&lt;h2 id="데드라인을-채우는-프리필-청크-sloweave--llm-서빙의-적응형-청킹-스케줄러"&gt;데드라인을 채우는 프리필 청크: SLOWeave — LLM 서빙의 적응형 청킹 스케줄러&lt;a href="#%eb%8d%b0%eb%93%9c%eb%9d%bc%ec%9d%b8%ec%9d%84-%ec%b1%84%ec%9a%b0%eb%8a%94-%ed%94%84%eb%a6%ac%ed%95%84-%ec%b2%ad%ed%81%ac-sloweave--llm-%ec%84%9c%eb%b9%99%ec%9d%98-%ec%a0%81%ec%9d%91%ed%98%95-%ec%b2%ad%ed%82%b9-%ec%8a%a4%ec%bc%80%ec%a4%84%eb%9f%ac" class="heading-anchor" aria-label="이 섹션에 대한 링크"&gt;&lt;svg class="h-4 w-4" aria-hidden="true" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"&gt;&lt;g fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2"&gt;&lt;path d="M10 13a5 5 0 0 0 7.54.54l3-3a5 5 0 0 0-7.07-7.07l-1.72 1.71"/&gt;&lt;path d="M14 11a5 5 0 0 0-7.54-.54l-3 3a5 5 0 0 0 7.07 7.07l1.71-1.71"/&gt;&lt;/g&gt;&lt;/svg&gt;&lt;/a&gt;&lt;/h2&gt;&#10; &lt;blockquote&#10; class="border-primary bg-muted/30 hover:bg-muted/40 my-6 rounded-r-lg border-l-4 py-4 pr-4 pl-6 italic transition-all duration-200 ease-out hover:translate-x-1"&#10; &gt;&#10; &lt;div class="prose prose-sm max-w-none"&gt;&#10; &lt;p&gt;&lt;strong&gt;TL;DR&lt;/strong&gt; — 고정된 프리필 청크 크기 대신, 매 반복(iteration)마다 &amp;ldquo;가장 이른 디코드 데드라인 전에 끝나는 가장 큰 프리필 청크&amp;quot;를 이진 탐색으로 선택하는 온라인 스케줄러 &lt;strong&gt;SLOWeave&lt;/strong&gt;를 제안한다. 25 ms TPOT 목표에서 혼합(mixed) 워크로드 goodput을 최강 고정 청크 대비 &lt;strong&gt;39%&lt;/strong&gt;, 장문(long) 워크로드에서 &lt;strong&gt;38%&lt;/strong&gt; 개선했고, 엄격한 10 ms 목표에서는 각각 &lt;strong&gt;3.3배&lt;/strong&gt;, &lt;strong&gt;2.4배&lt;/strong&gt;까지 끌어올린다 (근거: Abstract).&lt;/p&gt;</description></item></channel></rss>