Add code link for TetriServe (#371)

LRY89757 · web-flow · commit 2d46d57d2386 · 2026-03-31T11:59:40.000-04:00
* tetriserve asplos

* update bib

* update code and pdf

* selected artifact
diff --git a/source/_data/SymbioticLab.bib b/source/_data/SymbioticLab.bib
@@ -2150,6 +2150,7 @@ @InProceedings{tetriserve:asplos26
   publist_confkey = {ASPLOS'26},
   publist_link    = {paper || tetriserve-asplos26.pdf},
   publist_topic   = {Systems + AI},
+  publist_link    = {code || https://github.com/DiT-Serving/TetriServe},
   publist_abstract = {
   Diffusion Transformer (DiT) models excel at generating high-quality images through iterative denoising steps, but serving them under strict Service Level Objectives (SLOs) is challenging due to their high computational cost, particularly at larger resolutions. Existing serving systems use fixed-degree sequence parallelism, which is inefficient for heterogeneous workloads with mixed resolutions and deadlines, leading to poor GPU utilization and low SLO attainment. In this paper, we propose step-level sequence parallelism to dynamically adjust the degree of parallelism of individual requests according to their deadlines. We present TetriServe, a DiT serving system that implements this strategy for highly efficient image generation. Specifically, TetriServe introduces a novel round-based scheduling mechanism that improves SLO attainment by (1) discretizing time into fixed rounds to make deadline-aware scheduling tractable, (2) adapting parallelism at the step level and minimizing GPU hour consumption, and (3) jointly packing requests to minimize late completions. Extensive evaluation on state-of-the-art DiT models shows that TetriServe achieves up to 32% higher SLO attainment compared to existing solutions without degrading image quality.
   }
diff --git a/source/open-source/index.md b/source/open-source/index.md
@@ -59,6 +59,9 @@ sections:
 
   - title: "Selected Software Artifacts"
     items:
+      - name: "TetriServe"
+        github_url: "https://github.com/DiT-Serving/TetriServe"
+        description: "Efficient serving system for mixed DiT workloads."
       - name: "Venn"
         github_url: "https://github.com/SymbioticLab/Venn"
         description: "Resource management for collaborative learning jobs."
diff --git a/source/publications/files/tetriserve:asplos26/tetriserve-asplos26.pdf b/source/publications/files/tetriserve:asplos26/tetriserve-asplos26.pdf
diff --git a/source/publications/index.md b/source/publications/index.md
@@ -120,6 +120,7 @@ venues:
       name: The 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems
       date: 2026-03-22
       url: https://asplos-conference.org/2026/
+      acceptance: 15.71%
     - key: ASPLOS'23
       name: The 28th ACM International Conference on Architectural Support for Programming Languages and Operating Systems
       date: 2023-03-25

Original file line number	Diff line number	Diff line change
`@@ -2150,6 +2150,7 @@ @InProceedings{tetriserve:asplos26`
`2150`	`2150`	`publist_confkey = {ASPLOS'26},`
`2151`	`2151`	`publist_link = {paper \|\| tetriserve-asplos26.pdf},`
`2152`	`2152`	`publist_topic = {Systems + AI},`
	`2153`	`+ publist_link = {code \|\| https://github.com/DiT-Serving/TetriServe},`
`2153`	`2154`	`publist_abstract = {`
`2154`	`2155`	Diffusion Transformer (DiT) models excel at generating high-quality images through iterative denoising steps, but serving them under strict Service Level Objectives (SLOs) is challenging due to their high computational cost, particularly at larger resolutions. Existing serving systems use fixed-degree sequence parallelism, which is inefficient for heterogeneous workloads with mixed resolutions and deadlines, leading to poor GPU utilization and low SLO attainment. In this paper, we propose step-level sequence parallelism to dynamically adjust the degree of parallelism of individual requests according to their deadlines. We present TetriServe, a DiT serving system that implements this strategy for highly efficient image generation. Specifically, TetriServe introduces a novel round-based scheduling mechanism that improves SLO attainment by (1) discretizing time into fixed rounds to make deadline-aware scheduling tractable, (2) adapting parallelism at the step level and minimizing GPU hour consumption, and (3) jointly packing requests to minimize late completions. Extensive evaluation on state-of-the-art DiT models shows that TetriServe achieves up to 32% higher SLO attainment compared to existing solutions without degrading image quality.
`2155`	`2156`	`}`