{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,5]],"date-time":"2026-06-05T20:15:40Z","timestamp":1780690540366,"version":"3.54.1"},"reference-count":81,"publisher":"Institute of Electrical and Electronics Engineers (IEEE)","issue":"6","license":[{"start":{"date-parts":[[2026,6,1]],"date-time":"2026-06-01T00:00:00Z","timestamp":1780272000000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/ieeexplore.ieee.org\/Xplorehelp\/downloads\/license-information\/IEEE.html"},{"start":{"date-parts":[[2026,6,1]],"date-time":"2026-06-01T00:00:00Z","timestamp":1780272000000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2026,6,1]],"date-time":"2026-06-01T00:00:00Z","timestamp":1780272000000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"funder":[{"name":"National Key R&D Program of China","award":["2025YFE0204100"],"award-info":[{"award-number":["2025YFE0204100"]}]},{"name":"Science and Technology Development Fund of Macao","award":["0074\/2025\/AMJ"],"award-info":[{"award-number":["0074\/2025\/AMJ"]}]},{"DOI":"10.13039\/501100021171","name":"Basic and Applied Basic Research Foundation of Guangdong Province","doi-asserted-by":"publisher","award":["2023B1515130002"],"award-info":[{"award-number":["2023B1515130002"]}],"id":[{"id":"10.13039\/501100021171","id-type":"DOI","asserted-by":"publisher"}]},{"name":"Shenzhen Basic Research Program","award":["KJZD20230923113800001"],"award-info":[{"award-number":["KJZD20230923113800001"]}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Trans. Parallel Distrib. Syst."],"published-print":{"date-parts":[[2026,6]]},"DOI":"10.1109\/tpds.2026.3683122","type":"journal-article","created":{"date-parts":[[2026,4,13]],"date-time":"2026-04-13T19:38:07Z","timestamp":1776109087000},"page":"1442-1457","source":"Crossref","is-referenced-by-count":0,"title":["Workload-Adapted Resource Allocation for LLM Distributed Serving in Serverless Clusters"],"prefix":"10.1109","volume":"37","author":[{"ORCID":"https:\/\/orcid.org\/0000-0002-4809-9543","authenticated-orcid":false,"given":"Yanying","family":"Lin","sequence":"first","affiliation":[{"name":"Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0005-1776-4773","authenticated-orcid":false,"given":"Shijie","family":"Peng","sequence":"additional","affiliation":[{"name":"Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0009-5551-2584","authenticated-orcid":false,"given":"Yanbo","family":"Li","sequence":"additional","affiliation":[{"name":"Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-3064-5841","authenticated-orcid":false,"given":"Shutian","family":"Luo","sequence":"additional","affiliation":[{"name":"University of Virginia, Charlottesville, VA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-7548-6223","authenticated-orcid":false,"given":"Haiying","family":"Shen","sequence":"additional","affiliation":[{"name":"University of Virginia, Charlottesville, VA, USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-6133-407X","authenticated-orcid":false,"given":"Kejiang","family":"Ye","sequence":"additional","affiliation":[{"name":"Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-9480-0356","authenticated-orcid":false,"given":"Chengzhong","family":"Xu","sequence":"additional","affiliation":[{"name":"University of Macau, Macau, China"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","article-title":"LLaMA: Open and efficient foundation language models","author":"Touvron","year":"2023"},{"key":"ref2","first-page":"1877","article-title":"Language models are few-shot learners","volume-title":"Proc. 34th Int. Conf. Neural Inf. Process. Syst.","author":"Brown"},{"key":"ref3","article-title":"BERT: Pre-training of deep bidirectional transformers for language understanding","author":"Devlin","year":"2019"},{"key":"ref4","article-title":"PaLM 2 technical report","author":"Anil","year":"2023"},{"key":"ref5","article-title":"Scaling laws for neural language models","author":"Kaplan","year":"2020"},{"key":"ref6","doi-asserted-by":"publisher","DOI":"10.1109\/SC41405.2020.00024"},{"key":"ref7","doi-asserted-by":"publisher","DOI":"10.1145\/3492321.3519584"},{"key":"ref8","doi-asserted-by":"publisher","DOI":"10.1145\/3302424.3303953"},{"key":"ref9","first-page":"24829","article-title":"Piper: Multidimensional planner for DNN parallelization","volume-title":"Proc. 35th Int. Conf. Neural Inf. Process. Syst.","author":"Tarnawski"},{"key":"ref10","first-page":"663","article-title":"AlpaServe: Statistical multiplexing with model parallelism for deep learning serving","volume-title":"Proc. 17th USENIX Symp. Operating Syst. Des. Implementation","author":"Li"},{"key":"ref11","doi-asserted-by":"publisher","DOI":"10.1145\/3676641.3716261"},{"key":"ref12","first-page":"62557","article-title":"SGLang: Efficient execution of structured language model programs","volume-title":"Proc. 38th Int. Conf. Neural Inf. Process. Syst.","author":"Zheng"},{"key":"ref13","doi-asserted-by":"publisher","DOI":"10.1109\/IISWC63097.2024.00015"},{"key":"ref14","first-page":"521","article-title":"Orca: A distributed serving system for transformer-based generative models","volume-title":"Proc. 16th USENIX Symp. Operating Syst. Des. Implementation","author":"Yu"},{"key":"ref15","doi-asserted-by":"publisher","DOI":"10.1145\/3620665.3640411"},{"key":"ref16","article-title":"MuxServe: Flexible spatial-temporal multiplexing for multiple LLM serving","volume-title":"Proc. 41st Int. Conf. Mach. Learn.","author":"Duan"},{"key":"ref17","doi-asserted-by":"publisher","DOI":"10.1145\/3503222.3507778"},{"key":"ref18","doi-asserted-by":"publisher","DOI":"10.1145\/3689031.3696072"},{"key":"ref19","first-page":"787","article-title":"SHEPHERD: Serving DNNs in the wild","volume-title":"Proc. 20th USENIX Symp. Netw. Syst. Des. Implementation","author":"Zhang"},{"key":"ref20","doi-asserted-by":"publisher","DOI":"10.1145\/3437801.3441593"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.1609\/aaai.v38i16.29810"},{"key":"ref22","doi-asserted-by":"publisher","DOI":"10.1145\/3620665.3640411"},{"key":"ref23","first-page":"613","article-title":"Clipper: A low-latency online prediction serving system","volume-title":"Proc. 14th USENIX Symp. Netw. Syst. Des. Implementation","author":"Crankshaw"},{"key":"ref24","first-page":"1041","article-title":"Cocktail: A multidimensional optimization for model serving in cloud","volume-title":"Proc. 19th USENIX Symp. Netw. Syst. Des. Implementation","author":"Gunasekaran"},{"key":"ref25","doi-asserted-by":"publisher","DOI":"10.1145\/3419111.3421285"},{"key":"ref26","first-page":"1049","article-title":"MArk: Exploiting cloud services for cost-effective, SLO-aware machine learning inference serving","volume-title":"Proc. 2019 USENIX Annu. Tech. Conf.","author":"Zhang"},{"key":"ref27","first-page":"397","article-title":"INFaaS: Automated model-less inference serving","volume-title":"Proc. 2021 USENIX Annu. Tech. Conf.","volume":"2021","author":"Romero"},{"key":"ref28","doi-asserted-by":"publisher","DOI":"10.1145\/3419111.3421280"},{"key":"ref29","first-page":"205","article-title":"Serverless in the wild: Characterizing and optimizing the serverless workload at a large cloud provider","volume-title":"Proc. USENIX Annu. Tech. Conf.","author":"Shahrad"},{"key":"ref30","doi-asserted-by":"publisher","DOI":"10.1109\/TSC.2022.3166553"},{"key":"ref31","doi-asserted-by":"publisher","DOI":"10.1145\/3620678.3624783"},{"key":"ref32","doi-asserted-by":"publisher","DOI":"10.1145\/3579371.3589069"},{"key":"ref33","doi-asserted-by":"publisher","DOI":"10.1145\/3600006.3613155"},{"key":"ref34","doi-asserted-by":"publisher","DOI":"10.1145\/3620678.3624785"},{"key":"ref35","first-page":"1505","article-title":"Doing more with less: Orchestrating serverless applications without an orchestrator","volume-title":"Proc. 20th USENIX Symp. Netw. Syst. Des. Implementation","author":"Liu"},{"key":"ref36","doi-asserted-by":"publisher","DOI":"10.1145\/3567955.3567960"},{"key":"ref37","doi-asserted-by":"publisher","DOI":"10.1145\/3503222.3507721"},{"key":"ref38","doi-asserted-by":"publisher","DOI":"10.1145\/3492321.3519581"},{"key":"ref39","first-page":"217","article-title":"PilotFish: Harvesting free cycles of cloud gaming with deep learning training","volume-title":"Proc. USENIX Annu. Tech. Conf.","author":"Zhang"},{"key":"ref40","doi-asserted-by":"publisher","DOI":"10.1145\/3503222.3507725"},{"key":"ref41","first-page":"625","article-title":"Memory harvesting in multi-GPU systems with hierarchical unified virtual memory","volume-title":"Proc. USENIX Annu. Tech. Conf.","author":"Choi"},{"key":"ref42","doi-asserted-by":"publisher","DOI":"10.1145\/3503222.3507709"},{"key":"ref43","first-page":"473","article-title":"Tetris: Memory-efficient serverless inference through tensor sharing","volume-title":"Proc. USENIX Annu. Tech. Conf.","author":"Li"},{"key":"ref44","first-page":"1535","article-title":"Towards high-performance transactional stateful serverless workflows with affinity-aware leasing","volume-title":"Proc. USENIX Annu. Tech. Conf.","author":"Zhao"},{"key":"ref45","doi-asserted-by":"publisher","DOI":"10.1145\/3669940.3707285"},{"key":"ref46","first-page":"597","article-title":"Torpor: GPU-enabled serverless computing for low-latency, resource-efficient inference","volume-title":"Proc. USENIX Annu. Tech. Conf.","author":"Yu"},{"key":"ref47","first-page":"135","article-title":"ServerlessLLM: Low-latency serverless inference for large language models","volume-title":"Proc. 18th USENIX Symp. Operating Syst. Des. Implementation","author":"Fu"},{"key":"ref48","first-page":"57","article-title":"DeepServe: Serverless large language model serving at scale","volume-title":"Proc. 2025 USENIX Annu. Tech. Conf.","author":"Hu"},{"key":"ref49","first-page":"167","article-title":"Jolteon: Unleashing the promise of serverless for serverless workflows","volume-title":"Proc. 21st USENIX Symp. Netw. Syst. Des. Implementation","author":"Zhang"},{"key":"ref50","article-title":"CaraServe: CPU-assisted and rank-aware LoRA serving for generative LLM inference","author":"Li","year":"2024"},{"key":"ref51","doi-asserted-by":"publisher","DOI":"10.1145\/3689031.3696073"},{"key":"ref52","first-page":"199","article-title":"Fork in the road: Reflections and optimizations for cold start latency in production serverless systems","volume-title":"Proc. 19th USENIX Symp. Operating Syst. Des. Implementation","author":"Chai"},{"key":"ref53","article-title":"PARALLELGPUOS: A concurrent OS-level GPU checkpoint and restore system using validated speculation","author":"Huang","year":"2024"},{"key":"ref54","doi-asserted-by":"publisher","DOI":"10.1145\/3373376.3378512"},{"key":"ref55","first-page":"443","article-title":"Serving DNNs like clockwork: Performance predictability from the bottom up","volume-title":"Proc. 14th USENIX Symp. Operating Syst. Des. Implementation","author":"Gujarati"},{"key":"ref56","doi-asserted-by":"publisher","DOI":"10.1145\/3477132.3483580"},{"key":"ref57","doi-asserted-by":"publisher","DOI":"10.1145\/3318464.3389758"},{"key":"ref58","first-page":"995","article-title":"Beware of fragmentation: Scheduling GPU-sharing workloads with fragmentation gradient descent","volume-title":"Proc. 2023 USENIX Annu. Tech. Conf.","author":"Weng"},{"key":"ref59","doi-asserted-by":"publisher","DOI":"10.1145\/3542929.3563477"},{"key":"ref60","doi-asserted-by":"publisher","DOI":"10.1145\/3567955.3567964"},{"key":"ref61","first-page":"805","article-title":"FIRM: An intelligent fine-grained resource management framework for SLO-oriented microservices","volume-title":"Proc. 14th USENIX Symp. Operating Syst. Des. Implementation","author":"Qiu"},{"key":"ref62","doi-asserted-by":"publisher","DOI":"10.1109\/ISSRE5003.2020.00014"},{"key":"ref63","doi-asserted-by":"publisher","DOI":"10.1145\/3297858.3304005"},{"key":"ref64","article-title":"SuperServe: Fine-grained inference serving for unpredictable workloads","author":"Khare","year":"2023"},{"key":"ref65","doi-asserted-by":"publisher","DOI":"10.1145\/3552326.3587437"},{"key":"ref66","doi-asserted-by":"publisher","DOI":"10.1145\/3732941"},{"key":"ref67","article-title":"Efficient interactive LLM serving with proxy model-based sequence length prediction","author":"Qiu","year":"2024"},{"key":"ref68","doi-asserted-by":"publisher","DOI":"10.1145\/3575693.3575721"},{"key":"ref69","doi-asserted-by":"publisher","DOI":"10.1145\/3341301.3359646"},{"key":"ref70","doi-asserted-by":"publisher","DOI":"10.1145\/3581784.3607060"},{"key":"ref71","doi-asserted-by":"publisher","DOI":"10.1145\/3352460.3358296"},{"key":"ref72","article-title":"Robust speech recognition via large-scale weak supervision","volume-title":"Proc. 40th Int. Conf. Mach. Learn.","author":"Radford"},{"key":"ref73","doi-asserted-by":"publisher","DOI":"10.5244\/c.30.87"},{"key":"ref74","doi-asserted-by":"publisher","DOI":"10.1145\/3600006.3613165"},{"key":"ref75","first-page":"709","article-title":"Characterization of large language model development in the datacenter","volume-title":"Proc. 21st USENIX Symp. Netw. Syst. Des. Implementation","author":"Hu"},{"key":"ref76","first-page":"173","article-title":"Llumnix: Dynamic scheduling for large language model serving","volume-title":"Proc. 18th USENIX Symp. Operating Syst. Des. Implementation","author":"Sun"},{"key":"ref77","first-page":"289","article-title":"Themis: Fair and efficient GPU cluster scheduling","volume-title":"Proc. 17th USENIX Symp. Netw. Syst. Des. Implementation","author":"Mahajan"},{"key":"ref78","doi-asserted-by":"publisher","DOI":"10.1145\/3600006.3613175"},{"key":"ref79","first-page":"117","article-title":"Taming throughput-latency tradeoff in LLM inference with sarathi-serve","volume-title":"Proc. 18th USENIX Symp. Operating Syst. Des. Implementation","author":"Agrawal"},{"key":"ref80","first-page":"263","article-title":"GPREEMPT: GPU preemptive scheduling made general and efficient","volume-title":"Proc. USENIX Annu. Tech. Conf.","author":"Fan"},{"key":"ref81","first-page":"533","article-title":"AntMan: Dynamic scaling on gpu clusters for deep learning","volume-title":"Proc. 14th USENIX Symp. Operating Syst. Des. Implementation","author":"Xiao"}],"container-title":["IEEE Transactions on Parallel and Distributed Systems"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/71\/11483288\/11479931.pdf?arnumber=11479931","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,6,5]],"date-time":"2026-06-05T19:40:57Z","timestamp":1780688457000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11479931\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,6]]},"references-count":81,"journal-issue":{"issue":"6"},"URL":"https:\/\/doi.org\/10.1109\/tpds.2026.3683122","relation":{},"ISSN":["1045-9219","1558-2183","2161-9883"],"issn-type":[{"value":"1045-9219","type":"print"},{"value":"1558-2183","type":"electronic"},{"value":"2161-9883","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,6]]}}}