papers/hanzo-platform/sections/99-conclusion.tex

\section{Conclusion}

\label{sec:conclusion}

We have presented Hanzo Platform, a PaaS designed for AI-native applications that treats LLM inference, vector databases, GPU compute, and model serving as first-class platform primitives. The AI-aware build system reduces deployment time by 4.7x, the unified resource model simplifies infrastructure management, and the cost-optimized autoscaler reduces inference costs by 38\%. Evaluation against five competing platforms demonstrates superior performance across deployment time, cost, scaling, and developer experience. Production deployment with 2,800+ applications and 47M inference requests over 14 months validates the practical viability of the approach. Hanzo Platform is available at \texttt{platform.hanzo.ai}.

\bibliographystyle{plain}
\begin{thebibliography}{28}

\bibitem{baseten2022}
Baseten.
\newblock Baseten: The fastest way to deploy {ML} models.
\newblock \emph{Baseten Documentation}, 2022.

\bibitem{bentoml2022}
BentoML.
\newblock {BentoML}: The unified model serving framework.
\newblock \emph{GitHub Repository}, 2022.

\bibitem{dokku2013}
Dokku.
\newblock Dokku: The smallest {PaaS} implementation you've ever seen.
\newblock \emph{GitHub Repository}, 2013.

\bibitem{dokploy2024}
Dokploy.
\newblock Dokploy: Self-hosted platform as a service.
\newblock \emph{GitHub Repository}, 2024.

\bibitem{flyio2020}
Fly.io.
\newblock Fly.io: Run your full stack apps close to your users.
\newblock \emph{Fly.io Documentation}, 2020.

\bibitem{heroku2007}
Heroku.
\newblock Heroku: Cloud application platform.
\newblock \emph{Heroku Documentation}, 2007.

\bibitem{kserve2021}
KServe.
\newblock {KServe}: Highly scalable and standards based model inference platform.
\newblock \emph{KServe Documentation}, 2021.

\bibitem{kubesphere2020}
KubeSphere.
\newblock {KubeSphere}: The container platform tailored for {Kubernetes}.
\newblock \emph{KubeSphere Documentation}, 2020.

\bibitem{mlflow2018}
M.~Zaharia, A.~Chen, A.~Davidson, et~al.
\newblock Accelerating the machine learning lifecycle with {MLflow}.
\newblock \emph{IEEE Data Engineering Bulletin}, 41(4):39--45, 2018.

\bibitem{modal2023}
Modal.
\newblock Modal: End the struggle with cloud infrastructure.
\newblock \emph{Modal Documentation}, 2023.

\bibitem{railway2021}
Railway.
\newblock Railway: Instant deployments, effortless scale.
\newblock \emph{Railway Documentation}, 2021.

\bibitem{render2019}
Render.
\newblock Render: Cloud application hosting.
\newblock \emph{Render Documentation}, 2019.

\bibitem{replicate2022}
Replicate.
\newblock Replicate: Run and fine-tune open-source models.
\newblock \emph{Replicate Documentation}, 2022.

\bibitem{seldon2019}
Seldon.
\newblock Seldon core: An open source platform to deploy machine learning models.
\newblock \emph{Seldon Documentation}, 2019.

\bibitem{together2023}
Together~AI.
\newblock Together: Fast inference and fine-tuning.
\newblock \emph{Together Documentation}, 2023.

\bibitem{vercel2020}
Vercel.
\newblock Vercel: Develop. Preview. Ship.
\newblock \emph{Vercel Documentation}, 2020.

\bibitem{burns2016borg}
B.~Burns, B.~Grant, D.~Oppenheimer, E.~Brewer, and J.~Wilkes.
\newblock Borg, omega, and {Kubernetes}.
\newblock \emph{ACM Queue}, 14(1):70--93, 2016.

\bibitem{hindman2011mesos}
B.~Hindman, A.~Konwinski, M.~Zaharia, et~al.
\newblock Mesos: A platform for fine-grained resource sharing in the data center.
\newblock In \emph{NSDI}, 2011.

\bibitem{kwon2023vllm}
W.~Kwon, Z.~Li, S.~Zhuang, et~al.
\newblock Efficient memory management for large language model serving with {PagedAttention}.
\newblock In \emph{SOSP}, 2023.

\bibitem{zheng2024sglang}
L.~Zheng, L.~Yin, Z.~Xie, et~al.
\newblock {SGLang}: Efficient execution of structured language model programs.
\newblock \emph{arXiv preprint arXiv:2312.07104}, 2024.

\bibitem{sheng2023flexgen}
Y.~Sheng, L.~Zheng, B.~Yuan, et~al.
\newblock {FlexGen}: High-throughput generative inference of large language models with a single {GPU}.
\newblock In \emph{ICML}, 2023.

\bibitem{pope2023efficiently}
R.~Pope, S.~Douglas, A.~Chowdhery, et~al.
\newblock Efficiently scaling transformer inference.
\newblock In \emph{MLSys}, 2023.

\bibitem{aminabadi2022deepspeed}
R.~Y. Aminabadi, S.~Rajbhandari, M.~Zhang, et~al.
\newblock {DeepSpeed} inference: Enabling efficient inference of transformer models at unprecedented scale.
\newblock In \emph{SC22}, 2022.

\bibitem{patel2024splitwise}
P.~Patel, E.~Choukse, C.~Zhang, et~al.
\newblock Splitwise: Efficient generative {LLM} inference using phase splitting.
\newblock In \emph{ISCA}, 2024.

\bibitem{agrawal2024taming}
A.~Agrawal, N.~Kedia, A.~Panwar, et~al.
\newblock Taming throughput-latency tradeoff in {LLM} inference with {Sarathi-Serve}.
\newblock In \emph{OSDI}, 2024.

\end{thebibliography}