Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 2508.02520

Huawei Cloud Model-as-a-Service on the CloudMatrix384 SuperPod

Ao Xiao·Bangzheng He·Baoquan Zhang·Baoxing Huai·Bingji Wang·Bo Wang·Bo Xu·Boyi Hou·Chan Yang·Changhong Liu·Cheng Cui·Chenyu Zhu·Cong Feng·Daohui Wang·Dayun Lin·Duo Zhao·Fengshao Zou·Fu Wang·Gangqiang Zhang·Gengyuan Dan·Guanjie Chen·Guodong Guan·Guodong Yang·Haifeng Li·Haipei Zhu·Haley Li·Hao Feng·Hao Huang·Hao Xu·Hengrui Ma·Hengtao Fan·Hui Liu·Jia Li·Jiang Liu·Jiang Xu·Jie Meng·Jinhan Xin·Junhao Hu·Juwei Chen·Lan Yu·Lanxin Miao·Liang Liu·Linan Jing·Lu Zhou·Meina Han·Mingkun Deng·Mingyu Deng·Naitian Deng·Nizhong Lin·Peihan Zhao·Peng Pan·Pengfei Shen·Ping Li·Qi Zhang·Qian Wang·Qin ZhC Qingrong Xia·Qingyi Zhang·Qunchao Fu·Ren Guo·Ruimin Gao·Shaochun Li·Sheng Long·Shentian Li·Shining Wan·Shuai Shen·Shuangfu Zeng·Shuming Jing·Siqi Yang·Song Zhang·Tao Xu·Tianlin Du·Ting Chen·Wanxu Wu·Wei Jiang·Weinan Tong·Weiwei Chen·Wen Peng·Wenli Zhou·Wenquan Yang·Wenxin Liang·Xiang Liu·Xiaoli Zhou·Xin Jin·Xinyu Duan·Xu Li·Xu Zhang·Xusheng Chen·Yalong Shan·Yang Gan·Yao Lu·Yi Deng·Yi Zheng·Ying Xiong·Yingfei Zheng·Yiyun Zheng·Yizhou Shan·Yong Gao·Yong Zhang·Yongqiang Yang·Yuanjin Gong
View original source ↗Read source document ↗Export metadata ↓

Abstract

Scaled-out MoE LLMs and scaled-up SuperPods create new systems challenges for production Model-as-a-Service (MaaS), requiring disaggregation, low-latency communication, and decentralized serving. This report presents xDeepServe, the production serving system behind Huawei Cloud's MaaS offering on CloudMatrix384, a 48-server SuperPod with 384 Ascend 910C chips connected by a high-bandwidth UB fabric and global shared memory. It serves models including DeepSeek, Kimi, GLM, Qwen, and MiniMax, among others. xDeepServe is built around Transformerless, a disaggregated execution architecture that decomposes transformer inference into modular units -- attention, feedforward, and MoE -- and supports disaggregated Prefill-Decode and MoE-Attention deployments. To enable disaggregation, we develop XCCL, a memory-semantic communication layer providing microsecond-level point-to-point and scalable all-to-all primitives, and we extend FlowServe with decentralized DP groups and techniques to mitigate stragglers and synchronization variance. In a peak decoding configuration, xDeepServe reaches 2400 tokens/s per Ascend 910C chip at ~50ms time-per-output-token (TPOT).

Explore related subjects

cs.DC

Keep this discovery

BibTeXRIS

Ao Xiao, Bangzheng He, Baoquan Zhang, Baoxing Huai, Bingji Wang, Bo Wang, Bo Xu, Boyi Hou, Chan Yang, Changhong Liu, Cheng Cui, Chenyu Zhu, Cong Feng, Daohui Wang, Dayun Lin, Duo Zhao, Fengshao Zou, Fu Wang, Gangqiang Zhang, Gengyuan Dan, Guanjie Chen, Guodong Guan, Guodong Yang, Haifeng Li, Haipei Zhu, Haley Li, Hao Feng, Hao Huang, Hao Xu, Hengrui Ma, Hengtao Fan, Hui Liu, Jia Li, Jiang Liu, Jiang Xu, Jie Meng, Jinhan Xin, Junhao Hu, Juwei Chen, Lan Yu, Lanxin Miao, Liang Liu, Linan Jing, Lu Zhou, Meina Han, Mingkun Deng, Mingyu Deng, Naitian Deng, Nizhong Lin, Peihan Zhao, Peng Pan, Pengfei Shen, Ping Li, Qi Zhang, Qian Wang, Qin ZhC Qingrong Xia, Qingyi Zhang, Qunchao Fu, Ren Guo, Ruimin Gao, Shaochun Li, Sheng Long, Shentian Li, Shining Wan, Shuai Shen, Shuangfu Zeng, Shuming Jing, Siqi Yang, Song Zhang, Tao Xu, Tianlin Du, Ting Chen, Wanxu Wu, Wei Jiang, Weinan Tong, Weiwei Chen, Wen Peng, Wenli Zhou, Wenquan Yang, Wenxin Liang, Xiang Liu, Xiaoli Zhou, Xin Jin, Xinyu Duan, Xu Li, Xu Zhang, Xusheng Chen, Yalong Shan, Yang Gan, Yao Lu, Yi Deng, Yi Zheng, Ying Xiong, Yingfei Zheng, Yiyun Zheng, Yizhou Shan, Yong Gao, Yong Zhang, Yongqiang Yang, Yuanjin Gong. 2025-08-04. Huawei Cloud Model-as-a-Service on the CloudMatrix384 SuperPod. https://arxiv.org/abs/2508.02520

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Online Treasure Hunt in Vertex-Permuted Dynamic Rings

We study the problem of treasure hunt by a group of $k \geq 1$ agents in vertex-permuted dynamic rings (VP). In this model, the $n$ vertices remain on a ring but are permuted at each time step. We first show that treasure hunt is impossible for any $k \leq n-3$ agents, if there are no restrictions on the sequence of permutations used in the dynamic ring. We then study the $VP(\delta)$ setting, in which for every pair $i, j$ of vertices, the edge $(i, j)$ is guaranteed to appear within $\delta$ steps. We show that the class $VP(\delta)$ is feasible only for $\delta \geq \left\lceil \frac{n-1}{2}\right\rceil$. For the one-agent case, we show a tight bound of $\Theta(\delta n)$ on the worst-case search time as well as competitive ratio of any online algorithm for treasure hunt, provided $\delta \geq 2n$. We then give an optimal algorithm for $k$ agents, thereby showing that $k$ agents can obtain a speedup of $k$ on the worst-case search time. Finally, in the R-VP setting, in which in every step, the vertices are arranged as a ring according to a random permutation, we show that treasure hunt takes expected $\Theta(n)$ steps against an oblivious adversary and $\Theta(n \log n)$ steps against an adaptive adversary.

cs.DC↗
arXiv2026

The Computing Channel: How Modulation Programs the Airwaves

Distributed computing and distributed artificial intelligence require frequent exchanges of intermediate results, although many applications need only an aggregate rather than messages from individual devices. Conventional systems recover each message before computing the aggregate, whereas over-the-air computation (OAC) exploits simultaneous transmission to obtain it directly. However, dominant OAC implementations rely on analog signaling, creating a mismatch with finite-precision data and digital communication procedures. This article presents digital function-oriented communication, in which finite-alphabet symbol representations and receiver decisions are jointly designed so that multiple-access superposition encodes the desired function without recovering individual inputs. We introduce its computational-constellation principle, main design approaches, extensions, and implementation challenges. Federated edge learning illustrates how the framework can reduce user-dependent data-bearing resources while operating directly on quantized model updates.

cs.DC↗
arXiv2026

Can AI Remediate Backend Failures Safely? GuardedAct with Blast-Radius-Aware Sandboxing

Large Language Models (LLMs) have shown promising capabilities in generating remediation actions for microservice failures. However, directly executing AI-generated repair actions in production risks cascading collateral damage. We propose GuardedAct, a sandbox-first remediation framework that interposes a blast-radius-aware verification layer between the LLM action generator and the production environment. GuardedAct operates in four phases: (1) ingesting a diagnosis report together with the live system topology and recent telemetry, (2) prompting an LLM to produce a ranked list of candidate remediation actions, (3) simulating each action in a lightweight digital-twin sandbox that estimates the blast radius and assigns a risk label, and (4) enforcing a rollback-confidence gate that auto-executes only low-risk actions while escalating high-risk ones for human review. We evaluate GuardedAct on five fault scenarios injected into the DeathStarBench social-network application. Experimental results show that GuardedAct achieves an overall recovery rate of 87.4% while reducing collateral damage by 79.7% relative to direct LLM execution (from 25.6% to 5.2%), at the cost of a modest sandbox-induced increase in mean time to recovery (approximately 8 s). Ablation studies confirm that each component contributes meaningfully to the safety-speed trade-off.

cs.DC↗