{"id":{"repo_id":"uiuc","oai_identifier":"oai:www.ideals.illinois.edu:2142/108505"},"canonical_url":"https://search.dev.ndltd.org/etd/uiuc/oai:www.ideals.illinois.edu:2142/108505","repository":{"repo_id":"uiuc","name":"University of Illinois - Urbana-Champaign","base_url":"https://www.ideals.illinois.edu/oai-pmh"},"display":{"title":"Performance benchmarking, analysis, and optimization of deep learning inference","abstract":"The world sees a proliferation of deep learning (DL) models and their wide adoption in different application domains. This has made the performance benchmarking, understanding, and optimization of DL inference an increasingly pressing task for both hardware designers and system providers, as they would like to offer the best possible computing system to serve DL models with the desired latency, throughput, and energy requirements while maximizing resource utilization. However, DL faces the following challenges in performance engineering. Benchmarking — While there have been significant efforts to develop benchmark suites that evaluate widely used DL models, developing, maintaining, and running benchmarks takes a non-trivial amount of effort, and DL benchmarking has been hampered in part due to the lack of representative and up-to-date benchmarking suites. Performance Understanding — Understanding the performance of DL workloads is challenging as their characteristics depend on the interplay between the models, frameworks, system libraries, and the hardware (or the HW/SW stack). Existing profiling tools are disjoint, however, and only focus on profiling within a particular level of the stack. This largely limits the types of analysis that can be performed on model execution. Optimization Advising — The current DL optimization process is manual and ad-hoc that requires a lot of effort and expertise. Existing tools lack the highly desired abilities to characterize ideal performance, identify sources of inefficiency, and quantify the benefits of potential optimizations. Such deficiencies have led to slow DL characterization/optimization cycles that cannot keep up with the fast pace at which new DL innovations are introduced. Evaluation and Comparison — The current DL landscape is fast-paced and is rife with non-uniform models, hardware/software (HW/SW) stacks, but lacks a DL benchmarking platform to facilitate evaluation and comparison of DL innovations, be it models, frameworks, libraries, or hardware. Due to the lack of a benchmarking platform, the current practice of evaluating the benefits of proposed DL innovations is both arduous and error-prone — stifling the adoption of the innovations. This thesis addresses the above challenges in DL performance engineering. First we introduce DLBricks, a composable benchmark generation design that reduces the effort of developing, maintaining, and running DL benchmarks. DLBricks decomposes DL models into a set of unique runnable networks and constructs the original model’s performance using the performance of the generated benchmarks. Then, we present XSP, an across-stack profiling design that correlates profiles from different sources to obtain a holistic and hierarchical view of DL model execution. XSP innovatively leverages distributed tracing and accurately capture the profiles at each level of the HW/SW stack in spite of profiling overhead. Next, we propose Benanza, a systematic DL benchmarking and analysis design that guides researchers to potential optimization opportunities and assesses hypothetical execution scenarios on GPUs. Finally, we design MLModelScope, a consistent, reproducible, and scalable DL benchmarking platform to facilitate evaluation and comparison of DL innovations. This thesis also briefly discusses TrIMS, TOPS, and CommScope which are developed based on the needs observed from the performance benchmarking and optimization work to solve relevant problems in the DL domain.","abstract_html":"The world sees a proliferation of deep learning (DL) models and their wide adoption in different application domains. This has made the performance benchmarking, understanding, and optimization of DL inference an increasingly pressing task for both hardware designers and system providers, as they would like to offer the best possible computing system to serve DL models with the desired latency, throughput, and energy requirements while maximizing resource utilization. However, DL faces the following challenges in performance engineering. Benchmarking — While there have been significant efforts to develop benchmark suites that evaluate widely used DL models, developing, maintaining, and running benchmarks takes a non-trivial amount of effort, and DL benchmarking has been hampered in part due to the lack of representative and up-to-date benchmarking suites. Performance Understanding — Understanding the performance of DL workloads is challenging as their characteristics depend on the interplay between the models, frameworks, system libraries, and the hardware (or the HW/SW stack). Existing profiling tools are disjoint, however, and only focus on profiling within a particular level of the stack. This largely limits the types of analysis that can be performed on model execution. Optimization Advising — The current DL optimization process is manual and ad-hoc that requires a lot of effort and expertise. Existing tools lack the highly desired abilities to characterize ideal performance, identify sources of inefficiency, and quantify the benefits of potential optimizations. Such deficiencies have led to slow DL characterization/optimization cycles that cannot keep up with the fast pace at which new DL innovations are introduced. Evaluation and Comparison — The current DL landscape is fast-paced and is rife with non-uniform models, hardware/software (HW/SW) stacks, but lacks a DL benchmarking platform to facilitate evaluation and comparison of DL innovations, be it models, frameworks, libraries, or hardware. Due to the lack of a benchmarking platform, the current practice of evaluating the benefits of proposed DL innovations is both arduous and error-prone — stifling the adoption of the innovations. This thesis addresses the above challenges in DL performance engineering. First we introduce DLBricks, a composable benchmark generation design that reduces the effort of developing, maintaining, and running DL benchmarks. DLBricks decomposes DL models into a set of unique runnable networks and constructs the original model’s performance using the performance of the generated benchmarks. Then, we present XSP, an across-stack profiling design that correlates profiles from different sources to obtain a holistic and hierarchical view of DL model execution. XSP innovatively leverages distributed tracing and accurately capture the profiles at each level of the HW/SW stack in spite of profiling overhead. Next, we propose Benanza, a systematic DL benchmarking and analysis design that guides researchers to potential optimization opportunities and assesses hypothetical execution scenarios on GPUs. Finally, we design MLModelScope, a consistent, reproducible, and scalable DL benchmarking platform to facilitate evaluation and comparison of DL innovations. This thesis also briefly discusses TrIMS, TOPS, and CommScope which are developed based on the needs observed from the performance benchmarking and optimization work to solve relevant problems in the DL domain.","abstract_has_math":false,"creators":["Li, Cheng"],"institution":"University of Illinois at Urbana-Champaign","degree_name":"Ph.D.","degree_level":"Dissertation","degree_discipline":"Computer Science","degree_department":null,"school":null,"contributors":["Hwu, Wen-mei","Fletcher, Christopher","Padua, David","Tan, Wei"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2020,"date_issued":"2020-10-07T20:59:56Z","date_published":"2020-10-07T20:59:56Z","updated_at":"2026-07-22T22:24:48Z","subjects":["deep learning, machine learning, performance analysis, benchmarking, optimization"],"languages":["en"],"rights":["Copyright 2020 Cheng Li"],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"http://hdl.handle.net/2142/108505","outbound_label":"Handle","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Hwu, Wen-mei","Fletcher, Christopher","Padua, David","Tan, Wei"]},{"key":"dc:creator","label":"Author","values":["Li, Cheng"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2020-10-07T20:59:56Z","2020-07-16","2020-08"]},{"key":"dc:type","label":"Dc Type","values":["text","Thesis"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Computer Science"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Dissertation"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Ph.D."]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["University of Illinois at Urbana-Champaign"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["deep learning, machine learning, performance analysis, benchmarking, optimization"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["en"]},{"key":"dc:rights","label":"Dc Rights","values":["Copyright 2020 Cheng Li"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["http://hdl.handle.net/2142/108505"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["The world sees a proliferation of deep learning (DL) models and their wide adoption in different application domains. This has made the performance benchmarking, understanding, and optimization of DL inference an increasingly pressing task for both hardware designers and system providers, as they would like to offer the best possible computing system to serve DL models with the desired latency, throughput, and energy requirements while maximizing resource utilization. However, DL faces the following challenges in performance engineering. Benchmarking — While there have been significant efforts to develop benchmark suites that evaluate widely used DL models, developing, maintaining, and running benchmarks takes a non-trivial amount of effort, and DL benchmarking has been hampered in part due to the lack of representative and up-to-date benchmarking suites. Performance Understanding — Understanding the performance of DL workloads is challenging as their characteristics depend on the interplay between the models, frameworks, system libraries, and the hardware (or the HW/SW stack). Existing profiling tools are disjoint, however, and only focus on profiling within a particular level of the stack. This largely limits the types of analysis that can be performed on model execution. Optimization Advising — The current DL optimization process is manual and ad-hoc that requires a lot of effort and expertise. Existing tools lack the highly desired abilities to characterize ideal performance, identify sources of inefficiency, and quantify the benefits of potential optimizations. Such deficiencies have led to slow DL characterization/optimization cycles that cannot keep up with the fast pace at which new DL innovations are introduced. Evaluation and Comparison — The current DL landscape is fast-paced and is rife with non-uniform models, hardware/software (HW/SW) stacks, but lacks a DL benchmarking platform to facilitate evaluation and comparison of DL innovations, be it models, frameworks, libraries, or hardware. Due to the lack of a benchmarking platform, the current practice of evaluating the benefits of proposed DL innovations is both arduous and error-prone — stifling the adoption of the innovations. This thesis addresses the above challenges in DL performance engineering. First we introduce DLBricks, a composable benchmark generation design that reduces the effort of developing, maintaining, and running DL benchmarks. DLBricks decomposes DL models into a set of unique runnable networks and constructs the original model’s performance using the performance of the generated benchmarks. Then, we present XSP, an across-stack profiling design that correlates profiles from different sources to obtain a holistic and hierarchical view of DL model execution. XSP innovatively leverages distributed tracing and accurately capture the profiles at each level of the HW/SW stack in spite of profiling overhead. Next, we propose Benanza, a systematic DL benchmarking and analysis design that guides researchers to potential optimization opportunities and assesses hypothetical execution scenarios on GPUs. Finally, we design MLModelScope, a consistent, reproducible, and scalable DL benchmarking platform to facilitate evaluation and comparison of DL innovations. This thesis also briefly discusses TrIMS, TOPS, and CommScope which are developed based on the needs observed from the performance benchmarking and optimization work to solve relevant problems in the DL domain.","Submission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2020-10-02 without embargo terms","The student, Cheng Li, accepted the attached license on 2020-07-15 at 15:59.","The student, Cheng Li, submitted this Dissertation for approval on 2020-07-15 at 16:04.","This Dissertation was approved for publication on 2020-07-16 at 10:33.","DSpace SAF Submission Ingestion Package generated from Vireo submission #15644 on 2020-10-02 at 15:14:09","Made available in DSpace on 2020-10-07T20:59:56Z (GMT). No. of bitstreams: 2 LI-DISSERTATION-2020.pdf: 3617677 bytes, checksum: d06155c0af75ee9ba1cc4f1ec4c3c1a3 (MD5) LICENSE.txt: 4205 bytes, checksum: 4e77b85f0f681cc843b1324b613bf59a (MD5) Previous issue date: 2020-07-16"]},{"key":"dc:format","label":"Dc Format","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Performance benchmarking, analysis, and optimization of deep learning inference"]}]}],"canonical_facts":{"dc:contributor":["Hwu, Wen-mei","Fletcher, Christopher","Padua, David","Tan, Wei"],"dc:creator":["Li, Cheng"],"dc:date":["2020-10-07T20:59:56Z","2020-07-16","2020-08"],"dc:description":["The world sees a proliferation of deep learning (DL) models and their wide adoption in different application domains. This has made the performance benchmarking, understanding, and optimization of DL inference an increasingly pressing task for both hardware designers and system providers, as they would like to offer the best possible computing system to serve DL models with the desired latency, throughput, and energy requirements while maximizing resource utilization. However, DL faces the following challenges in performance engineering. Benchmarking — While there have been significant efforts to develop benchmark suites that evaluate widely used DL models, developing, maintaining, and running benchmarks takes a non-trivial amount of effort, and DL benchmarking has been hampered in part due to the lack of representative and up-to-date benchmarking suites. Performance Understanding — Understanding the performance of DL workloads is challenging as their characteristics depend on the interplay between the models, frameworks, system libraries, and the hardware (or the HW/SW stack). Existing profiling tools are disjoint, however, and only focus on profiling within a particular level of the stack. This largely limits the types of analysis that can be performed on model execution. Optimization Advising — The current DL optimization process is manual and ad-hoc that requires a lot of effort and expertise. Existing tools lack the highly desired abilities to characterize ideal performance, identify sources of inefficiency, and quantify the benefits of potential optimizations. Such deficiencies have led to slow DL characterization/optimization cycles that cannot keep up with the fast pace at which new DL innovations are introduced. Evaluation and Comparison — The current DL landscape is fast-paced and is rife with non-uniform models, hardware/software (HW/SW) stacks, but lacks a DL benchmarking platform to facilitate evaluation and comparison of DL innovations, be it models, frameworks, libraries, or hardware. Due to the lack of a benchmarking platform, the current practice of evaluating the benefits of proposed DL innovations is both arduous and error-prone — stifling the adoption of the innovations. This thesis addresses the above challenges in DL performance engineering. First we introduce DLBricks, a composable benchmark generation design that reduces the effort of developing, maintaining, and running DL benchmarks. DLBricks decomposes DL models into a set of unique runnable networks and constructs the original model’s performance using the performance of the generated benchmarks. Then, we present XSP, an across-stack profiling design that correlates profiles from different sources to obtain a holistic and hierarchical view of DL model execution. XSP innovatively leverages distributed tracing and accurately capture the profiles at each level of the HW/SW stack in spite of profiling overhead. Next, we propose Benanza, a systematic DL benchmarking and analysis design that guides researchers to potential optimization opportunities and assesses hypothetical execution scenarios on GPUs. Finally, we design MLModelScope, a consistent, reproducible, and scalable DL benchmarking platform to facilitate evaluation and comparison of DL innovations. This thesis also briefly discusses TrIMS, TOPS, and CommScope which are developed based on the needs observed from the performance benchmarking and optimization work to solve relevant problems in the DL domain.","Submission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2020-10-02 without embargo terms","The student, Cheng Li, accepted the attached license on 2020-07-15 at 15:59.","The student, Cheng Li, submitted this Dissertation for approval on 2020-07-15 at 16:04.","This Dissertation was approved for publication on 2020-07-16 at 10:33.","DSpace SAF Submission Ingestion Package generated from Vireo submission #15644 on 2020-10-02 at 15:14:09","Made available in DSpace on 2020-10-07T20:59:56Z (GMT). No. of bitstreams: 2 LI-DISSERTATION-2020.pdf: 3617677 bytes, checksum: d06155c0af75ee9ba1cc4f1ec4c3c1a3 (MD5) LICENSE.txt: 4205 bytes, checksum: 4e77b85f0f681cc843b1324b613bf59a (MD5) Previous issue date: 2020-07-16"],"dc:format":["application/pdf"],"dc:identifier":["http://hdl.handle.net/2142/108505"],"dc:language":["en"],"dc:rights":["Copyright 2020 Cheng Li"],"dc:subject":["deep learning, machine learning, performance analysis, benchmarking, optimization"],"dc:title":["Performance benchmarking, analysis, and optimization of deep learning inference"],"dc:type":["text","Thesis"],"thesis:degree_discipline":["Computer Science"],"thesis:degree_level":["Dissertation"],"thesis:degree_name":["Ph.D."],"thesis:institution_name":["University of Illinois at Urbana-Champaign"]},"updated_at":"2026-07-22T22:24:48Z"}