{"id":{"repo_id":"gmu","oai_identifier":"oai:MARS:1920/13717"},"canonical_url":"https://search.dev.ndltd.org/etd/gmu/oai:MARS:1920/13717","repository":{"repo_id":"gmu","name":"George Mason University","base_url":"https://mars.gmu.edu/server/oai/request"},"display":{"title":"Towards Elastic and Cost-effective Stateful Serverless Systems","abstract":"Serverless computing has emerged as a new paradigm that transforms conventional VM-based cloud computing by offering fine-grained computational resources with high elasticity while abstracting away complex system administration tasks. This computing model has garnered significant interest due to its autoscaling capabilities, massive parallelism, and fine-grained pay-per-use billing. Function-as-a-Service (FaaS) is a state-of-the-art example of the serverless service, which provides developers with cloud functions running relatively small pieces of code written in a high-level language to construct stateless serverless applications. In this dissertation, we develop a series of solutions to extend serverless computingbenefits to stateful systems. Our research starts with utilizing an existing stateless FaaS offering to create a stateful in-memory cache. To address the statelessness of cloud functions, we study function behavior. FaaS providers often cache functions to mitigate cold start penalties. Our six-month-long study reveals that this caching policy can be used for retaining data through multiplexed periodic function invocations. By integrating more fault-tolerant techniques such as erasure coding and periodic delta backup, we introduce the InfiniCache, a first-of-its-kind, cost-effective in-memory object cache based on stateless serverless functions. It is the first step towards building stateful serverless systems. The InfiniCache gains 31 − 96× tenant-side cost savings compared to AWS ElastiCachefor a large-object-only production workload, maintains an effective 95.4% data durability for each one hour window, and offers competitive performance comparable to typical in-memory caches. Next, we present InfiniStore, a serverless cloud storage with the ServerlessMemoryservice. Building upon our initial solution, InfiniCache, ServerlessMemory employs a sliding-window-based memory management strategy, inspired by the garbage collection mechanisms in programming languages, to effectively segregate hot and cold data. This approach provides fine-grained elasticity, good performance, and an extremely low-cost pay-per-access cost model. Combined with an inexpensive cloud object layer, InfiniStore ensures durability despite function failures through a fast parallel recovery scheme built on the auto-scaling functionality of a FaaS platform. The fine-grained elasticity achieved results of 26.26% cost benefits compared to our initial solution. Lastly, we propose InfiniMLT, a GPU-enabled stateful serverless service for OnlineInteractive Machine Learning Training Service (OIMLTS), designed to optimize task interactivity, GPU utilization, and cost. All three solutions demonstrate the feasibility of leveraging serverless computing for stateful applications while maintaining key properties such as elasticity, cost-effectiveness, and pay-per-use.","abstract_html":"Serverless computing has emerged as a new paradigm that transforms conventional VM-based cloud computing by offering fine-grained computational resources with high elasticity while abstracting away complex system administration tasks. This computing model has garnered significant interest due to its autoscaling capabilities, massive parallelism, and fine-grained pay-per-use billing. Function-as-a-Service (FaaS) is a state-of-the-art example of the serverless service, which provides developers with cloud functions running relatively small pieces of code written in a high-level language to construct stateless serverless applications. In this dissertation, we develop a series of solutions to extend serverless computingbenefits to stateful systems. Our research starts with utilizing an existing stateless FaaS offering to create a stateful in-memory cache. To address the statelessness of cloud functions, we study function behavior. FaaS providers often cache functions to mitigate cold start penalties. Our six-month-long study reveals that this caching policy can be used for retaining data through multiplexed periodic function invocations. By integrating more fault-tolerant techniques such as erasure coding and periodic delta backup, we introduce the InfiniCache, a first-of-its-kind, cost-effective in-memory object cache based on stateless serverless functions. It is the first step towards building stateful serverless systems. The InfiniCache gains 31 − 96× tenant-side cost savings compared to AWS ElastiCachefor a large-object-only production workload, maintains an effective 95.4% data durability for each one hour window, and offers competitive performance comparable to typical in-memory caches. Next, we present InfiniStore, a serverless cloud storage with the ServerlessMemoryservice. Building upon our initial solution, InfiniCache, ServerlessMemory employs a sliding-window-based memory management strategy, inspired by the garbage collection mechanisms in programming languages, to effectively segregate hot and cold data. This approach provides fine-grained elasticity, good performance, and an extremely low-cost pay-per-access cost model. Combined with an inexpensive cloud object layer, InfiniStore ensures durability despite function failures through a fast parallel recovery scheme built on the auto-scaling functionality of a FaaS platform. The fine-grained elasticity achieved results of 26.26% cost benefits compared to our initial solution. Lastly, we propose InfiniMLT, a GPU-enabled stateful serverless service for OnlineInteractive Machine Learning Training Service (OIMLTS), designed to optimize task interactivity, GPU utilization, and cost. All three solutions demonstrate the feasibility of leveraging serverless computing for stateful applications while maintaining key properties such as elasticity, cost-effectiveness, and pay-per-use.","abstract_has_math":false,"creators":["Zhang, Jingyuan"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2023,"date_issued":"2023","date_published":"2023","updated_at":"2026-07-27T19:51:46Z","subjects":["Cost-effective","Elastic","GPU","Serverless","Stateful","Storage"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["hdl:1920/13717"],"render_values":[{"text":"hdl:1920/13717","href":null,"code":true}]}]},"links":{"outbound_url":null,"outbound_label":null,"outbound_source":null},"metadata_groups":[{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.issued","label":"Date","values":["2023"]},{"key":"dc:type","label":"Dc Type","values":["Dissertation"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Cost-effective","Elastic","GPU","Serverless","Stateful","Storage"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["hdl:1920/13717"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.other","label":"Dc Description Other","values":["Serverless computing has emerged as a new paradigm that transforms conventional VM-based cloud computing by offering fine-grained computational resources with high elasticity while abstracting away complex system administration tasks. This computing model has garnered significant interest due to its autoscaling capabilities, massive parallelism, and fine-grained pay-per-use billing. Function-as-a-Service (FaaS) is a state-of-the-art example of the serverless service, which provides developers with cloud functions running relatively small pieces of code written in a high-level language to construct stateless serverless applications. In this dissertation, we develop a series of solutions to extend serverless computingbenefits to stateful systems. Our research starts with utilizing an existing stateless FaaS offering to create a stateful in-memory cache. To address the statelessness of cloud functions, we study function behavior. FaaS providers often cache functions to mitigate cold start penalties. Our six-month-long study reveals that this caching policy can be used for retaining data through multiplexed periodic function invocations. By integrating more fault-tolerant techniques such as erasure coding and periodic delta backup, we introduce the InfiniCache, a first-of-its-kind, cost-effective in-memory object cache based on stateless serverless functions. It is the first step towards building stateful serverless systems. The InfiniCache gains 31 − 96× tenant-side cost savings compared to AWS ElastiCachefor a large-object-only production workload, maintains an effective 95.4% data durability for each one hour window, and offers competitive performance comparable to typical in-memory caches. Next, we present InfiniStore, a serverless cloud storage with the ServerlessMemoryservice. Building upon our initial solution, InfiniCache, ServerlessMemory employs a sliding-window-based memory management strategy, inspired by the garbage collection mechanisms in programming languages, to effectively segregate hot and cold data. This approach provides fine-grained elasticity, good performance, and an extremely low-cost pay-per-access cost model. Combined with an inexpensive cloud object layer, InfiniStore ensures durability despite function failures through a fast parallel recovery scheme built on the auto-scaling functionality of a FaaS platform. The fine-grained elasticity achieved results of 26.26% cost benefits compared to our initial solution. Lastly, we propose InfiniMLT, a GPU-enabled stateful serverless service for OnlineInteractive Machine Learning Training Service (OIMLTS), designed to optimize task interactivity, GPU utilization, and cost. All three solutions demonstrate the feasibility of leveraging serverless computing for stateful applications while maintaining key properties such as elasticity, cost-effectiveness, and pay-per-use."]},{"key":"dc:title","label":"Title","values":["Towards Elastic and Cost-effective Stateful Serverless Systems"]}]}],"canonical_facts":{"dc:date.issued":["2023"],"dc:description.other":["Serverless computing has emerged as a new paradigm that transforms conventional VM-based cloud computing by offering fine-grained computational resources with high elasticity while abstracting away complex system administration tasks. This computing model has garnered significant interest due to its autoscaling capabilities, massive parallelism, and fine-grained pay-per-use billing. Function-as-a-Service (FaaS) is a state-of-the-art example of the serverless service, which provides developers with cloud functions running relatively small pieces of code written in a high-level language to construct stateless serverless applications. In this dissertation, we develop a series of solutions to extend serverless computingbenefits to stateful systems. Our research starts with utilizing an existing stateless FaaS offering to create a stateful in-memory cache. To address the statelessness of cloud functions, we study function behavior. FaaS providers often cache functions to mitigate cold start penalties. Our six-month-long study reveals that this caching policy can be used for retaining data through multiplexed periodic function invocations. By integrating more fault-tolerant techniques such as erasure coding and periodic delta backup, we introduce the InfiniCache, a first-of-its-kind, cost-effective in-memory object cache based on stateless serverless functions. It is the first step towards building stateful serverless systems. The InfiniCache gains 31 − 96× tenant-side cost savings compared to AWS ElastiCachefor a large-object-only production workload, maintains an effective 95.4% data durability for each one hour window, and offers competitive performance comparable to typical in-memory caches. Next, we present InfiniStore, a serverless cloud storage with the ServerlessMemoryservice. Building upon our initial solution, InfiniCache, ServerlessMemory employs a sliding-window-based memory management strategy, inspired by the garbage collection mechanisms in programming languages, to effectively segregate hot and cold data. This approach provides fine-grained elasticity, good performance, and an extremely low-cost pay-per-access cost model. Combined with an inexpensive cloud object layer, InfiniStore ensures durability despite function failures through a fast parallel recovery scheme built on the auto-scaling functionality of a FaaS platform. The fine-grained elasticity achieved results of 26.26% cost benefits compared to our initial solution. Lastly, we propose InfiniMLT, a GPU-enabled stateful serverless service for OnlineInteractive Machine Learning Training Service (OIMLTS), designed to optimize task interactivity, GPU utilization, and cost. All three solutions demonstrate the feasibility of leveraging serverless computing for stateful applications while maintaining key properties such as elasticity, cost-effectiveness, and pay-per-use."],"dc:identifier":["hdl:1920/13717"],"dc:subject":["Cost-effective","Elastic","GPU","Serverless","Stateful","Storage"],"dc:title":["Towards Elastic and Cost-effective Stateful Serverless Systems"],"dc:type":["Dissertation"]},"updated_at":"2026-07-27T19:51:46Z"}