_version_ 1866916358859522048
author Jin, Yibo
Wang, Tao
Lin, Huimin
Song, Mingyang
Li, Peiyang
Ma, Yipeng
Shan, Yicheng
Yuan, Zhengfan
Li, Cailong
Sun, Yajing
Wu, Tiandeng
Chu, Xing
Huan, Ruizhi
Ma, Li
You, Xiao
Zhou, Wenting
Ye, Yunpeng
Liu, Wen
Xu, Xiangkun
Zhang, Yongsheng
Dong, Tiantian
Zhu, Jiawei
Wang, Zhe
Ju, Xijian
Song, Jianxun
Cheng, Haoliang
Li, Xiaojing
Ding, Jiandong
Guo, Hefei
Zhang, Zhengyong
author_facet Jin, Yibo
Wang, Tao
Lin, Huimin
Song, Mingyang
Li, Peiyang
Ma, Yipeng
Shan, Yicheng
Yuan, Zhengfan
Li, Cailong
Sun, Yajing
Wu, Tiandeng
Chu, Xing
Huan, Ruizhi
Ma, Li
You, Xiao
Zhou, Wenting
Ye, Yunpeng
Liu, Wen
Xu, Xiangkun
Zhang, Yongsheng
Dong, Tiantian
Zhu, Jiawei
Wang, Zhe
Ju, Xijian
Song, Jianxun
Cheng, Haoliang
Li, Xiaojing
Ding, Jiandong
Guo, Hefei
Zhang, Zhengyong
contents Serving disaggregated large language models (LLMs) over tens of thousands of xPU devices (GPUs or NPUs) with reliable performance faces multiple challenges. 1) Ignoring the diversity (various prefixes and tidal requests), treating all the prompts in a mixed pool is inadequate. To facilitate the similarity per scenario and minimize the inner mismatch on P/D (prefill and decoding) processing, fine-grained organization is required, dynamically adjusting P/D ratios for better performance. 2) Due to inaccurate estimation on workload (queue status or maintained connections), the global scheduler easily incurs unnecessary timeouts in prefill. 3) Block-fixed device-to-device (D2D) KVCache transfer over cluster-level RDMA (remote direct memory access) fails to achieve desired D2D utilization as expected. To overcome previous problems, this paper proposes an end-to-end system P/D-Serve, complying with the paradigm of MLOps (machine learning operations), which models end-to-end (E2E) P/D performance and enables: 1) fine-grained P/D organization, mapping the service with RoCE (RDMA over converged ethernet) as needed, to facilitate similar processing and dynamic adjustments on P/D ratios; 2) on-demand forwarding upon rejections for idle prefill, decoupling the scheduler from regular inaccurate reports and local queues, to avoid timeouts in prefill; and 3) efficient KVCache transfer via optimized D2D access. P/D-Serve is implemented upon Ascend and MindSpore, has been deployed over tens of thousands of NPUs for more than eight months in commercial use, and further achieves 60\%, 42\% and 46\% improvements on E2E throughput, time-to-first-token (TTFT) SLO (service level objective) and D2D transfer time. As the E2E system with optimizations, P/D-Serve achieves 6.7x increase on throughput, compared with aggregated LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2408_08147
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle P/D-Serve: Serving Disaggregated Large Language Model at Scale
Jin, Yibo
Wang, Tao
Lin, Huimin
Song, Mingyang
Li, Peiyang
Ma, Yipeng
Shan, Yicheng
Yuan, Zhengfan
Li, Cailong
Sun, Yajing
Wu, Tiandeng
Chu, Xing
Huan, Ruizhi
Ma, Li
You, Xiao
Zhou, Wenting
Ye, Yunpeng
Liu, Wen
Xu, Xiangkun
Zhang, Yongsheng
Dong, Tiantian
Zhu, Jiawei
Wang, Zhe
Ju, Xijian
Song, Jianxun
Cheng, Haoliang
Li, Xiaojing
Ding, Jiandong
Guo, Hefei
Zhang, Zhengyong
Distributed, Parallel, and Cluster Computing
Computation and Language
Machine Learning
Serving disaggregated large language models (LLMs) over tens of thousands of xPU devices (GPUs or NPUs) with reliable performance faces multiple challenges. 1) Ignoring the diversity (various prefixes and tidal requests), treating all the prompts in a mixed pool is inadequate. To facilitate the similarity per scenario and minimize the inner mismatch on P/D (prefill and decoding) processing, fine-grained organization is required, dynamically adjusting P/D ratios for better performance. 2) Due to inaccurate estimation on workload (queue status or maintained connections), the global scheduler easily incurs unnecessary timeouts in prefill. 3) Block-fixed device-to-device (D2D) KVCache transfer over cluster-level RDMA (remote direct memory access) fails to achieve desired D2D utilization as expected. To overcome previous problems, this paper proposes an end-to-end system P/D-Serve, complying with the paradigm of MLOps (machine learning operations), which models end-to-end (E2E) P/D performance and enables: 1) fine-grained P/D organization, mapping the service with RoCE (RDMA over converged ethernet) as needed, to facilitate similar processing and dynamic adjustments on P/D ratios; 2) on-demand forwarding upon rejections for idle prefill, decoupling the scheduler from regular inaccurate reports and local queues, to avoid timeouts in prefill; and 3) efficient KVCache transfer via optimized D2D access. P/D-Serve is implemented upon Ascend and MindSpore, has been deployed over tens of thousands of NPUs for more than eight months in commercial use, and further achieves 60\%, 42\% and 46\% improvements on E2E throughput, time-to-first-token (TTFT) SLO (service level objective) and D2D transfer time. As the E2E system with optimizations, P/D-Serve achieves 6.7x increase on throughput, compared with aggregated LLMs.
title P/D-Serve: Serving Disaggregated Large Language Model at Scale
topic Distributed, Parallel, and Cluster Computing
Computation and Language
Machine Learning
url https://arxiv.org/abs/2408.08147