{"id":{"repo_id":"cornell","oai_identifier":"oai:ecommons.cornell.edu:1813/116641"},"canonical_url":"https://search.ndltd.org/etd/cornell/oai:ecommons.cornell.edu:1813/116641","repository":{"repo_id":"cornell","name":"Cornell University","base_url":"https://ecommons.cornell.edu/server/oai/request"},"display":{"title":"Optimizing Foundational System Building Blocks of Datacenter Applications","abstract":"Cloud computing has become the prevailing computing infrastructure for the majority of the world's computation. Computing platforms for cloud computing and large internet services are hosted in datacenters, and optimizing the performance of datacenter applications can result in significant cost savings. Given the diversity of datacenter workloads, optimizing a single application may not yield substantial improvements in the total system efficiency, as costs are spread across numerous independent workloads. In contrast, optimizing the foundational system building blocks of datacenter applications, including high-level system infrastructures to underlying system software libraries, can significantly improve the productivity of the datacenter fleet, since entire classes of datacenter applications can benefit from such optimizations. This dissertation proposes a series of optimizations in foundational system building blocks of datacenter applications. Applications running in datacenter are often built as collections of loosely coupled services that are deployed and executed through high-level system building blocks such as serverless workflow engines and microservice frameworks. First, we focus on optimizing such a system building block at the top of the computing stack, the serverless computing framework. Despite the benefits of ease of programming, fast elasticity, and fine-grained billing, serverless computing suffers from resource inefficiency. We designed Aquatope, a QoS-and-uncertainty-aware resource scheduler for end-to-end serverless workflows that takes into account the inherent uncertainty present in FaaS platforms, and improves performance predictability and resource efficiency. Aquatope uses a set of scalable and validated Bayesian models to create prewarmed containers ahead of function invocations, and to allocate appropriate resources at function granularity to meet a complex workflow’s end-to-end QoS, while minimizing resource cost. Aquatope demonstrates that a joint solution to cold start and resource management, taking into account uncertainty, can effectively improve the resource efficiency of serverless applications. However, serverless workflows still suffer from significant control plane and inter-function communication overheads, which make them unsuitable for latency-critical applications. We also designed Meteion, a fast and efficient serverless workflow engine for latency-critical interactive applications. Meteion decouples the control plane from the workflow execution, and leverages lightweight per-function engines to enable decentralized workflow orchestration and direct inter-function communication. Meteion's DAG scheduler utilizes the workflow's latency distribution and graph structure to provision containers promptly, ensuring that functions can execute seamlessly on worker servers without falling back to the control plane. Second, we delve into a foundational system library, the memory allocator. Datacenter applications typically share the usage of certain low-level software libraries, and memory allocation constitutes a substantial component of datacenter computation. Optimizing the memory allocator can improve application performance, leading to significant cost savings. We present the first comprehensive characterization of TCMalloc at warehouse scale. Our characterization reveals a profound diversity in the memory allocation patterns, allocated object sizes and lifetimes, for large-scale datacenter workloads, as well as in their performance on heterogeneous hardware platforms. Based on these insights, we optimize TCMalloc for warehouse-scale environments. Specifically, we propose optimizations for each level of its cache hierarchy that include usage-based dynamic sizing of allocator caches, leveraging hardware topology to mitigate inter-core communication overhead, and improving allocation packing algorithms based on statistical data. Evaluation results show that these optimizations significantly improve the productivity of the datacenter fleet.","abstract_html":"Cloud computing has become the prevailing computing infrastructure for the majority of the world&#x27;s computation. Computing platforms for cloud computing and large internet services are hosted in datacenters, and optimizing the performance of datacenter applications can result in significant cost savings. Given the diversity of datacenter workloads, optimizing a single application may not yield substantial improvements in the total system efficiency, as costs are spread across numerous independent workloads. In contrast, optimizing the foundational system building blocks of datacenter applications, including high-level system infrastructures to underlying system software libraries, can significantly improve the productivity of the datacenter fleet, since entire classes of datacenter applications can benefit from such optimizations. This dissertation proposes a series of optimizations in foundational system building blocks of datacenter applications. Applications running in datacenter are often built as collections of loosely coupled services that are deployed and executed through high-level system building blocks such as serverless workflow engines and microservice frameworks. First, we focus on optimizing such a system building block at the top of the computing stack, the serverless computing framework. Despite the benefits of ease of programming, fast elasticity, and fine-grained billing, serverless computing suffers from resource inefficiency. We designed Aquatope, a QoS-and-uncertainty-aware resource scheduler for end-to-end serverless workflows that takes into account the inherent uncertainty present in FaaS platforms, and improves performance predictability and resource efficiency. Aquatope uses a set of scalable and validated Bayesian models to create prewarmed containers ahead of function invocations, and to allocate appropriate resources at function granularity to meet a complex workflow’s end-to-end QoS, while minimizing resource cost. Aquatope demonstrates that a joint solution to cold start and resource management, taking into account uncertainty, can effectively improve the resource efficiency of serverless applications. However, serverless workflows still suffer from significant control plane and inter-function communication overheads, which make them unsuitable for latency-critical applications. We also designed Meteion, a fast and efficient serverless workflow engine for latency-critical interactive applications. Meteion decouples the control plane from the workflow execution, and leverages lightweight per-function engines to enable decentralized workflow orchestration and direct inter-function communication. Meteion&#x27;s DAG scheduler utilizes the workflow&#x27;s latency distribution and graph structure to provision containers promptly, ensuring that functions can execute seamlessly on worker servers without falling back to the control plane. Second, we delve into a foundational system library, the memory allocator. Datacenter applications typically share the usage of certain low-level software libraries, and memory allocation constitutes a substantial component of datacenter computation. Optimizing the memory allocator can improve application performance, leading to significant cost savings. We present the first comprehensive characterization of TCMalloc at warehouse scale. Our characterization reveals a profound diversity in the memory allocation patterns, allocated object sizes and lifetimes, for large-scale datacenter workloads, as well as in their performance on heterogeneous hardware platforms. Based on these insights, we optimize TCMalloc for warehouse-scale environments. Specifically, we propose optimizations for each level of its cache hierarchy that include usage-based dynamic sizing of allocator caches, leveraging hardware topology to mitigate inter-core communication overhead, and improving allocation packing algorithms based on statistical data. Evaluation results show that these optimizations significantly improve the productivity of the datacenter fleet.","abstract_has_math":false,"creators":["Zhou, Zhuangzhuang"],"institution":"Cornell University","degree_name":"Ph. D., Electrical and Computer Engineering","degree_level":"Doctor of Philosophy","degree_discipline":"Electrical and Computer Engineering","degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":["Delimitrou, Christina"],"committee_members":["Martinez, Jose","Weatherspoon, Hakim"],"year":2024,"date_issued":"2024-08","date_published":"2024-08","updated_at":"2026-11-19T03:12:03Z","subjects":["Cloud Computing","Datacenter","Memory Allocator","Memory Management","Resource Management","Serverless Computing"],"languages":["en"],"rights":[],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier.doi","label":"DOI","values":["https://doi.org/10.7298/rhcv-wh67"],"render_values":[{"text":"https://doi.org/10.7298/rhcv-wh67","href":"https://doi.org/10.7298/rhcv-wh67","code":true}]},{"key":"dc:identifier.bibid","label":"Dc Identifier Bibid","values":["16611851"],"render_values":[{"text":"16611851","href":null,"code":true}]},{"key":"dc:identifier.other","label":"Dc Identifier Other","values":["ProQuest Submission ID: 14444","ProQuest Publication ID: 31482554"],"render_values":[{"text":"ProQuest Submission ID: 14444","href":null,"code":true},{"text":"ProQuest Publication ID: 31482554","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/1813/116641","outbound_label":"Handle","outbound_source":"dc:identifier.uri","record_url":"https://hdl.handle.net/1813/116641","record_label":"Repository record","record_kind":"handle","record_source":"dc:identifier.uri","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://ecommons.cornell.edu/server/oai/request?verb=GetRecord&metadataPrefix=dim&identifier=oai%3Aecommons.cornell.edu%3A1813%2F116641","prefix":"dim"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.chair","label":"Chair","values":["Delimitrou, Christina"]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Martinez, Jose","Weatherspoon, Hakim"]},{"key":"dc:creator","label":"Author","values":["Zhou, Zhuangzhuang"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2025-01-14T20:01:26Z"]},{"key":"dc:date.issued","label":"Date","values":["2024-08"]},{"key":"dc:relation.uri","label":"Dc Relation URI","values":["https://newcatalog.library.cornell.edu/catalog/16611851"]},{"key":"dc:type","label":"Dc Type","values":["dissertation or thesis"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Electrical and Computer Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Doctor of Philosophy"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Ph. D., Electrical and Computer Engineering"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["Cornell University"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Cloud Computing","Datacenter","Memory Allocator","Memory Management","Resource Management","Serverless Computing"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.bibid","label":"Dc Identifier Bibid","values":["16611851"]},{"key":"dc:identifier.doi","label":"DOI","values":["https://doi.org/10.7298/rhcv-wh67"]},{"key":"dc:identifier.other","label":"Dc Identifier Other","values":["ProQuest Submission ID: 14444","ProQuest Publication ID: 31482554"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/1813/116641"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"cerif:openaire","label":"Cerif Openaire","values":["<Publication xmlns=\"https://www.openaire.eu/cerif-profile/1.1/\" id=\"b72020de-7279-4471-be77-2c7b45001097\"> <Type xmlns=\"https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types\">http://purl.org/coar/resource_type/c_1843</Type> <Language>en</Language> <Title>Optimizing Foundational System Building Blocks of Datacenter Applications</Title> <PublishedIn> <Publication> </Publication> </PublishedIn> <PublicationDate>2024-08</PublicationDate> <DOI>https://doi.org/10.7298/rhcv-wh67</DOI> <Authors> <Author> <DisplayName>Zhou, Zhuangzhuang</DisplayName> <Affiliation> <OrgUnit> </OrgUnit> </Affiliation> </Author> </Authors> <Editors> </Editors> <Publishers> <Publisher> <OrgUnit /> </Publisher> </Publishers> <Keyword>Cloud Computing</Keyword> <Keyword>Datacenter</Keyword> <Keyword>Memory Allocator</Keyword> <Keyword>Memory Management</Keyword> <Keyword>Resource Management</Keyword> <Keyword>Serverless Computing</Keyword> <Abstract>Cloud computing has become the prevailing computing infrastructure for the majority of the world's computation. Computing platforms for cloud computing and large internet services are hosted in datacenters, and optimizing the performance of datacenter applications can result in significant cost savings. Given the diversity of datacenter workloads, optimizing a single application may not yield substantial improvements in the total system efficiency, as costs are spread across numerous independent workloads. In contrast, optimizing the foundational system building blocks of datacenter applications, including high-level system infrastructures to underlying system software libraries, can significantly improve the productivity of the datacenter fleet, since entire classes of datacenter applications can benefit from such optimizations. This dissertation proposes a series of optimizations in foundational system building blocks of datacenter applications. Applications running in datacenter are often built as collections of loosely coupled services that are deployed and executed through high-level system building blocks such as serverless workflow engines and microservice frameworks. First, we focus on optimizing such a system building block at the top of the computing stack, the serverless computing framework. Despite the benefits of ease of programming, fast elasticity, and fine-grained billing, serverless computing suffers from resource inefficiency. We designed Aquatope, a QoS-and-uncertainty-aware resource scheduler for end-to-end serverless workflows that takes into account the inherent uncertainty present in FaaS platforms, and improves performance predictability and resource efficiency. Aquatope uses a set of scalable and validated Bayesian models to create prewarmed containers ahead of function invocations, and to allocate appropriate resources at function granularity to meet a complex workflow’s end-to-end QoS, while minimizing resource cost. Aquatope demonstrates that a joint solution to cold start and resource management, taking into account uncertainty, can effectively improve the resource efficiency of serverless applications. However, serverless workflows still suffer from significant control plane and inter-function communication overheads, which make them unsuitable for latency-critical applications. We also designed Meteion, a fast and efficient serverless workflow engine for latency-critical interactive applications. Meteion decouples the control plane from the workflow execution, and leverages lightweight per-function engines to enable decentralized workflow orchestration and direct inter-function communication. Meteion's DAG scheduler utilizes the workflow's latency distribution and graph structure to provision containers promptly, ensuring that functions can execute seamlessly on worker servers without falling back to the control plane. Second, we delve into a foundational system library, the memory allocator. Datacenter applications typically share the usage of certain low-level software libraries, and memory allocation constitutes a substantial component of datacenter computation. Optimizing the memory allocator can improve application performance, leading to significant cost savings. We present the first comprehensive characterization of TCMalloc at warehouse scale. Our characterization reveals a profound diversity in the memory allocation patterns, allocated object sizes and lifetimes, for large-scale datacenter workloads, as well as in their performance on heterogeneous hardware platforms. Based on these insights, we optimize TCMalloc for warehouse-scale environments. Specifically, we propose optimizations for each level of its cache hierarchy that include usage-based dynamic sizing of allocator caches, leveraging hardware topology to mitigate inter-core communication overhead, and improving allocation packing algorithms based on statistical data. Evaluation results show that these optimizations significantly improve the productivity of the datacenter fleet.</Abstract> <Access xmlns=\"http://purl.org/coar/access_right\" > </Access> </Publication>"]},{"key":"cris:virtual.author","label":"Cris Virtual Author","values":["Zhou, Zhuangzhuang"]},{"key":"cris:virtualsource.collection","label":"Cris Virtualsource Collection","values":["5893a6ea-7af3-41d7-abc6-04bcd26ab5df"]},{"key":"dc:description","label":"Description","values":["158 pages"]},{"key":"dc:description.abstract","label":"Abstract","values":["Cloud computing has become the prevailing computing infrastructure for the majority of the world's computation. Computing platforms for cloud computing and large internet services are hosted in datacenters, and optimizing the performance of datacenter applications can result in significant cost savings. Given the diversity of datacenter workloads, optimizing a single application may not yield substantial improvements in the total system efficiency, as costs are spread across numerous independent workloads. In contrast, optimizing the foundational system building blocks of datacenter applications, including high-level system infrastructures to underlying system software libraries, can significantly improve the productivity of the datacenter fleet, since entire classes of datacenter applications can benefit from such optimizations. This dissertation proposes a series of optimizations in foundational system building blocks of datacenter applications. Applications running in datacenter are often built as collections of loosely coupled services that are deployed and executed through high-level system building blocks such as serverless workflow engines and microservice frameworks. First, we focus on optimizing such a system building block at the top of the computing stack, the serverless computing framework. Despite the benefits of ease of programming, fast elasticity, and fine-grained billing, serverless computing suffers from resource inefficiency. We designed Aquatope, a QoS-and-uncertainty-aware resource scheduler for end-to-end serverless workflows that takes into account the inherent uncertainty present in FaaS platforms, and improves performance predictability and resource efficiency. Aquatope uses a set of scalable and validated Bayesian models to create prewarmed containers ahead of function invocations, and to allocate appropriate resources at function granularity to meet a complex workflow’s end-to-end QoS, while minimizing resource cost. Aquatope demonstrates that a joint solution to cold start and resource management, taking into account uncertainty, can effectively improve the resource efficiency of serverless applications. However, serverless workflows still suffer from significant control plane and inter-function communication overheads, which make them unsuitable for latency-critical applications. We also designed Meteion, a fast and efficient serverless workflow engine for latency-critical interactive applications. Meteion decouples the control plane from the workflow execution, and leverages lightweight per-function engines to enable decentralized workflow orchestration and direct inter-function communication. Meteion's DAG scheduler utilizes the workflow's latency distribution and graph structure to provision containers promptly, ensuring that functions can execute seamlessly on worker servers without falling back to the control plane. Second, we delve into a foundational system library, the memory allocator. Datacenter applications typically share the usage of certain low-level software libraries, and memory allocation constitutes a substantial component of datacenter computation. Optimizing the memory allocator can improve application performance, leading to significant cost savings. We present the first comprehensive characterization of TCMalloc at warehouse scale. Our characterization reveals a profound diversity in the memory allocation patterns, allocated object sizes and lifetimes, for large-scale datacenter workloads, as well as in their performance on heterogeneous hardware platforms. Based on these insights, we optimize TCMalloc for warehouse-scale environments. Specifically, we propose optimizations for each level of its cache hierarchy that include usage-based dynamic sizing of allocator caches, leveraging hardware topology to mitigate inter-core communication overhead, and improving allocation packing algorithms based on statistical data. Evaluation results show that these optimizations significantly improve the productivity of the datacenter fleet."]},{"key":"dc:description.collection","label":"Dc Description Collection","values":["Cornell Theses and Dissertations"]},{"key":"dc:format.mimetype","label":"Dc Format Mimetype","values":["application/pdf"]},{"key":"dcterms:license","label":"Dcterms License","values":["https://hdl.handle.net/1813/59810.2"]},{"key":"dc:title","label":"Title","values":["Optimizing Foundational System Building Blocks of Datacenter Applications"]},{"key":"dspace:entity.type","label":"Dspace Entity Type","values":["Publication"]},{"key":"others:access_status","label":"Others Access Status","values":["open.access"]}]}],"canonical_facts":{"cerif:openaire":["<Publication xmlns=\"https://www.openaire.eu/cerif-profile/1.1/\" id=\"b72020de-7279-4471-be77-2c7b45001097\"> <Type xmlns=\"https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types\">http://purl.org/coar/resource_type/c_1843</Type> <Language>en</Language> <Title>Optimizing Foundational System Building Blocks of Datacenter Applications</Title> <PublishedIn> <Publication> </Publication> </PublishedIn> <PublicationDate>2024-08</PublicationDate> <DOI>https://doi.org/10.7298/rhcv-wh67</DOI> <Authors> <Author> <DisplayName>Zhou, Zhuangzhuang</DisplayName> <Affiliation> <OrgUnit> </OrgUnit> </Affiliation> </Author> </Authors> <Editors> </Editors> <Publishers> <Publisher> <OrgUnit /> </Publisher> </Publishers> <Keyword>Cloud Computing</Keyword> <Keyword>Datacenter</Keyword> <Keyword>Memory Allocator</Keyword> <Keyword>Memory Management</Keyword> <Keyword>Resource Management</Keyword> <Keyword>Serverless Computing</Keyword> <Abstract>Cloud computing has become the prevailing computing infrastructure for the majority of the world's computation. Computing platforms for cloud computing and large internet services are hosted in datacenters, and optimizing the performance of datacenter applications can result in significant cost savings. Given the diversity of datacenter workloads, optimizing a single application may not yield substantial improvements in the total system efficiency, as costs are spread across numerous independent workloads. In contrast, optimizing the foundational system building blocks of datacenter applications, including high-level system infrastructures to underlying system software libraries, can significantly improve the productivity of the datacenter fleet, since entire classes of datacenter applications can benefit from such optimizations. This dissertation proposes a series of optimizations in foundational system building blocks of datacenter applications. Applications running in datacenter are often built as collections of loosely coupled services that are deployed and executed through high-level system building blocks such as serverless workflow engines and microservice frameworks. First, we focus on optimizing such a system building block at the top of the computing stack, the serverless computing framework. Despite the benefits of ease of programming, fast elasticity, and fine-grained billing, serverless computing suffers from resource inefficiency. We designed Aquatope, a QoS-and-uncertainty-aware resource scheduler for end-to-end serverless workflows that takes into account the inherent uncertainty present in FaaS platforms, and improves performance predictability and resource efficiency. Aquatope uses a set of scalable and validated Bayesian models to create prewarmed containers ahead of function invocations, and to allocate appropriate resources at function granularity to meet a complex workflow’s end-to-end QoS, while minimizing resource cost. Aquatope demonstrates that a joint solution to cold start and resource management, taking into account uncertainty, can effectively improve the resource efficiency of serverless applications. However, serverless workflows still suffer from significant control plane and inter-function communication overheads, which make them unsuitable for latency-critical applications. We also designed Meteion, a fast and efficient serverless workflow engine for latency-critical interactive applications. Meteion decouples the control plane from the workflow execution, and leverages lightweight per-function engines to enable decentralized workflow orchestration and direct inter-function communication. Meteion's DAG scheduler utilizes the workflow's latency distribution and graph structure to provision containers promptly, ensuring that functions can execute seamlessly on worker servers without falling back to the control plane. Second, we delve into a foundational system library, the memory allocator. Datacenter applications typically share the usage of certain low-level software libraries, and memory allocation constitutes a substantial component of datacenter computation. Optimizing the memory allocator can improve application performance, leading to significant cost savings. We present the first comprehensive characterization of TCMalloc at warehouse scale. Our characterization reveals a profound diversity in the memory allocation patterns, allocated object sizes and lifetimes, for large-scale datacenter workloads, as well as in their performance on heterogeneous hardware platforms. Based on these insights, we optimize TCMalloc for warehouse-scale environments. Specifically, we propose optimizations for each level of its cache hierarchy that include usage-based dynamic sizing of allocator caches, leveraging hardware topology to mitigate inter-core communication overhead, and improving allocation packing algorithms based on statistical data. Evaluation results show that these optimizations significantly improve the productivity of the datacenter fleet.</Abstract> <Access xmlns=\"http://purl.org/coar/access_right\" > </Access> </Publication>"],"cris:virtual.author":["Zhou, Zhuangzhuang"],"cris:virtualsource.collection":["5893a6ea-7af3-41d7-abc6-04bcd26ab5df"],"dc:contributor.chair":["Delimitrou, Christina"],"dc:contributor.committeemember":["Martinez, Jose","Weatherspoon, Hakim"],"dc:creator":["Zhou, Zhuangzhuang"],"dc:date.accessioned":["2025-01-14T20:01:26Z"],"dc:date.issued":["2024-08"],"dc:description":["158 pages"],"dc:description.abstract":["Cloud computing has become the prevailing computing infrastructure for the majority of the world's computation. Computing platforms for cloud computing and large internet services are hosted in datacenters, and optimizing the performance of datacenter applications can result in significant cost savings. Given the diversity of datacenter workloads, optimizing a single application may not yield substantial improvements in the total system efficiency, as costs are spread across numerous independent workloads. In contrast, optimizing the foundational system building blocks of datacenter applications, including high-level system infrastructures to underlying system software libraries, can significantly improve the productivity of the datacenter fleet, since entire classes of datacenter applications can benefit from such optimizations. This dissertation proposes a series of optimizations in foundational system building blocks of datacenter applications. Applications running in datacenter are often built as collections of loosely coupled services that are deployed and executed through high-level system building blocks such as serverless workflow engines and microservice frameworks. First, we focus on optimizing such a system building block at the top of the computing stack, the serverless computing framework. Despite the benefits of ease of programming, fast elasticity, and fine-grained billing, serverless computing suffers from resource inefficiency. We designed Aquatope, a QoS-and-uncertainty-aware resource scheduler for end-to-end serverless workflows that takes into account the inherent uncertainty present in FaaS platforms, and improves performance predictability and resource efficiency. Aquatope uses a set of scalable and validated Bayesian models to create prewarmed containers ahead of function invocations, and to allocate appropriate resources at function granularity to meet a complex workflow’s end-to-end QoS, while minimizing resource cost. Aquatope demonstrates that a joint solution to cold start and resource management, taking into account uncertainty, can effectively improve the resource efficiency of serverless applications. However, serverless workflows still suffer from significant control plane and inter-function communication overheads, which make them unsuitable for latency-critical applications. We also designed Meteion, a fast and efficient serverless workflow engine for latency-critical interactive applications. Meteion decouples the control plane from the workflow execution, and leverages lightweight per-function engines to enable decentralized workflow orchestration and direct inter-function communication. Meteion's DAG scheduler utilizes the workflow's latency distribution and graph structure to provision containers promptly, ensuring that functions can execute seamlessly on worker servers without falling back to the control plane. Second, we delve into a foundational system library, the memory allocator. Datacenter applications typically share the usage of certain low-level software libraries, and memory allocation constitutes a substantial component of datacenter computation. Optimizing the memory allocator can improve application performance, leading to significant cost savings. We present the first comprehensive characterization of TCMalloc at warehouse scale. Our characterization reveals a profound diversity in the memory allocation patterns, allocated object sizes and lifetimes, for large-scale datacenter workloads, as well as in their performance on heterogeneous hardware platforms. Based on these insights, we optimize TCMalloc for warehouse-scale environments. Specifically, we propose optimizations for each level of its cache hierarchy that include usage-based dynamic sizing of allocator caches, leveraging hardware topology to mitigate inter-core communication overhead, and improving allocation packing algorithms based on statistical data. Evaluation results show that these optimizations significantly improve the productivity of the datacenter fleet."],"dc:description.collection":["Cornell Theses and Dissertations"],"dc:format.mimetype":["application/pdf"],"dc:identifier.bibid":["16611851"],"dc:identifier.doi":["https://doi.org/10.7298/rhcv-wh67"],"dc:identifier.other":["ProQuest Submission ID: 14444","ProQuest Publication ID: 31482554"],"dc:identifier.uri":["https://hdl.handle.net/1813/116641"],"dc:language.iso":["en"],"dc:relation.uri":["https://newcatalog.library.cornell.edu/catalog/16611851"],"dc:subject":["Cloud Computing","Datacenter","Memory Allocator","Memory Management","Resource Management","Serverless Computing"],"dcterms:license":["https://hdl.handle.net/1813/59810.2"],"dc:title":["Optimizing Foundational System Building Blocks of Datacenter Applications"],"dc:type":["dissertation or thesis"],"dspace:entity.type":["Publication"],"others:access_status":["open.access"],"thesis:degree_discipline":["Electrical and Computer Engineering"],"thesis:degree_level":["Doctor of Philosophy"],"thesis:degree_name":["Ph. D., Electrical and Computer Engineering"],"thesis:institution_name":["Cornell University"]},"updated_at":"2026-11-19T03:12:03Z"}