國立虎尾科技大學 |

Cost Effective Machine Learning for Computer Architecture Design.

紀錄類型:	書目-語言資料,手稿 : Monograph/item
正題名/作者:	Cost Effective Machine Learning for Computer Architecture Design./
作者:	Amiraski, Maziar.
面頁冊數:	1 online resource (110 pages)
附註:	Source: Dissertations Abstracts International, Volume: 85-11, Section: A.
Contained By:	Dissertations Abstracts International85-11A.
標題:	Computer engineering. -
電子資源:	click for full text (PQDT)
ISBN:	9798382737515

Cost Effective Machine Learning for Computer Architecture Design.
Amiraski, Maziar.

Cost Effective Machine Learning for Computer Architecture Design. - 1 online resource (110 pages)

Source: Dissertations Abstracts International, Volume: 85-11, Section: A.

Thesis (Ph.D.)--Tufts University, 2024.

Includes bibliographical references

The computer architecture design landscape is faced by considerable challenges, including the growing diversity and complexity of workloads, the increasing heterogeneity of hardware platforms, and the need to balance competing objectives such as performance, power efficiency, and reliability. The sheer size of the design space and the difficulty of optimizing across multiple dimensions present significant obstacles. Machine learning has shown the potential to address these problems and drastically change the course of computer architecture design. First, traditional heuristic approaches to system design often struggle to cope with the increasing complexity of modern workloads and hardware platforms. Machine learning techniques offer a more flexible and adaptive alternative, capable of extracting insights from vast datasets to optimize architecture parameters. Moreover, machine learning enables a shift towards more holistic optimization strategies, considering multiple objectives simultaneously and accommodating diverse design constraints.Utilizing an accurate and low cost machine learning model to design effective computer systems is the goal of this dissertation. We showcase two machine learning models for two different purposes: In the first part, we present how we designed a highly accurate classifier with low overhead to partition the last level cache (LLC). LLC being a shared resource, is faced with performance and fairness challenges when running competing workloads with diverse behaviors. CASHT demonstrates my research focus on characterizing systems under contention and collecting data for the purpose of training machine learning models. These models, while accurate, have low hardware implementation overhead (considering the area, memory and latency) and are specifically designed to be integrated into current standard platforms.In the second part, we address the pressing issue of managing advanced hotspots on modern microprocessors, emphasizing its detrimental effect on performance, product reliability, and device lifetime which is only exacerbated as we decrease the transistor sizes. We introduce Boreas, an effective control system based on a low-cost regression model to prevent hotspots on single digit process nodes. Boreas demonstrates another dimension of my research on the use of machine learning for system design by training a model to assist the DVFS controller. Boreas outperforms existing thermal management techniques while remaining lightweight and well-suited for implementation in hardware.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2024

Mode of access: World Wide Web

ISBN: 9798382737515Subjects--Topical Terms:

569006
Computer engineering.
Subjects--Index Terms:

Machine learningIndex Terms--Genre/Form:

554714
Electronic books.

Cost Effective Machine Learning for Computer Architecture Design.
LDR:03946ntm a22004097 4500 001 1149737
005 20241022112627.5
006 m o d
007 cr bn ---uuuuu
008 250605s2024 xx obm 000 0 eng d
020 $a 9798382737515
035 $a (MiAaPQ)AAI31293336
035 $a AAI31293336
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Amiraski, Maziar. $3 1476067
245 1 0 $a Cost Effective Machine Learning for Computer Architecture Design.
264 0 $c 2024
300 $a 1 online resource (110 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 85-11, Section: A.
500 $a Advisor: Hempstead, Mark.
502 $a Thesis (Ph.D.)--Tufts University, 2024.
504 $a Includes bibliographical references
520 $a The computer architecture design landscape is faced by considerable challenges, including the growing diversity and complexity of workloads, the increasing heterogeneity of hardware platforms, and the need to balance competing objectives such as performance, power efficiency, and reliability. The sheer size of the design space and the difficulty of optimizing across multiple dimensions present significant obstacles. Machine learning has shown the potential to address these problems and drastically change the course of computer architecture design. First, traditional heuristic approaches to system design often struggle to cope with the increasing complexity of modern workloads and hardware platforms. Machine learning techniques offer a more flexible and adaptive alternative, capable of extracting insights from vast datasets to optimize architecture parameters. Moreover, machine learning enables a shift towards more holistic optimization strategies, considering multiple objectives simultaneously and accommodating diverse design constraints.Utilizing an accurate and low cost machine learning model to design effective computer systems is the goal of this dissertation. We showcase two machine learning models for two different purposes: In the first part, we present how we designed a highly accurate classifier with low overhead to partition the last level cache (LLC). LLC being a shared resource, is faced with performance and fairness challenges when running competing workloads with diverse behaviors. CASHT demonstrates my research focus on characterizing systems under contention and collecting data for the purpose of training machine learning models. These models, while accurate, have low hardware implementation overhead (considering the area, memory and latency) and are specifically designed to be integrated into current standard platforms.In the second part, we address the pressing issue of managing advanced hotspots on modern microprocessors, emphasizing its detrimental effect on performance, product reliability, and device lifetime which is only exacerbated as we decrease the transistor sizes. We introduce Boreas, an effective control system based on a low-cost regression model to prevent hotspots on single digit process nodes. Boreas demonstrates another dimension of my research on the use of machine learning for system design by training a model to assist the DVFS controller. Boreas outperforms existing thermal management techniques while remaining lightweight and well-suited for implementation in hardware.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Computer engineering. $3 569006
650 4 $a Architectural engineering. $3 1180400
650 4 $a Landscape architecture. $3 555495
653 $a Machine learning
653 $a Last level cache
653 $a Computer architecture design
653 $a Hardware platforms
653 $a Microprocessors
655 7 $a Electronic books. $2 local $3 554714
690 $a 0464
690 $a 0800
690 $a 0390
690 $a 0462
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a Tufts University. $b Electrical Engineering. $3 1184138
773 0 $t Dissertations Abstracts International $g 85-11A.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=31293336 $z click for full text (PQDT)