東華大學圖書館 |

語系: 繁體中文

說明(常見問題)

回圖書館首頁

手機版館藏查詢

登入

回首頁

切換: 標籤 | MARC模式 | ISBD

FindBook

Google Book

Amazon

博客來

In-Database Machine Learning on Reconfigurable Dataflow Accelerators.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	In-Database Machine Learning on Reconfigurable Dataflow Accelerators./
作者:	Vilim, Matthew.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2021,
面頁冊數:	120 p.
附註:	Source: Dissertations Abstracts International, Volume: 83-03, Section: A.
Contained By:	Dissertations Abstracts International83-03A.
標題:	Flexibility. -
電子資源:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28688324
ISBN:	9798544204022

In-Database Machine Learning on Reconfigurable Dataflow Accelerators.
Vilim, Matthew.

In-Database Machine Learning on Reconfigurable Dataflow Accelerators. - Ann Arbor : ProQuest Dissertations & Theses, 2021 - 120 p.

Source: Dissertations Abstracts International, Volume: 83-03, Section: A.

Thesis (Ph.D.)--Stanford University, 2021.

This item must not be sold to any third party vendors.

Stagnant cpu performance is driving an explosion in domain-specific architectures to supplement cpus for data-intensive workloads in data centers [1, 2, 3, 4]. These accelerators sacrifice generality and programmability beyond their target domain in exchange for higher-performance [5, 1].However, accelerator deployment in data centers remains limited outside all but the most ubiquitous of application domains like machine learning [5]. High-performance systems require large dice in advanced process nodes with many supporting resources like high-bandwidth memory. As a result, to justify the price that comes with highperformance ic design, designers building non-ml accelerators are left with a choice: either piggyback on ml accelerators-giving up some efficiency but taking advantage of supporting hardware resources-or build more-efficient but less-advanced bespoke hardware.In this work, we show that reconfigurable dataflow accelerators (rdas) are a practical alternative to building fixed-function designs per application domain. We extend Plasticine [5]-a previously proposed ml-focused rda- with low-overhead, microarchitectural extensions to support analytic database queries. These extensions increase area by just 4 %, and the unified accelerator outperforms a multi-core software baseline by 1500 x.Finally, we show how to re-purpose these extensions to implement data structures like trees and hash tables that are critical to asymptotically optimal query plans. We introduce a threading model for vector dataflow accelerators that extracts parallelism from data structures with irregular control flow using fine-grained thread scheduling- outperforming a gpu by 8x.

ISBN: 9798544204022Subjects--Topical Terms:

3560705
Flexibility.

In-Database Machine Learning on Reconfigurable Dataflow Accelerators.
LDR:02763nmm a2200325 4500 001 2344862
005 20220531062200.5
008 241004s2021 ||||||||||||||||| ||eng d
020 $a 9798544204022
035 $a (MiAaPQ)AAI28688324
035 $a (MiAaPQ)STANFORDbr114bf8393
035 $a AAI28688324
040 $a MiAaPQ $c MiAaPQ
100 1 $a Vilim, Matthew. $3 3683688
245 1 0 $a In-Database Machine Learning on Reconfigurable Dataflow Accelerators.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2021
300 $a 120 p.
500 $a Source: Dissertations Abstracts International, Volume: 83-03, Section: A.
500 $a Advisor: Olukotun, Oyekunle Ayinde; Hennessy, John L.; Re, Christopher.
502 $a Thesis (Ph.D.)--Stanford University, 2021.
506 $a This item must not be sold to any third party vendors.
520 $a Stagnant cpu performance is driving an explosion in domain-specific architectures to supplement cpus for data-intensive workloads in data centers [1, 2, 3, 4]. These accelerators sacrifice generality and programmability beyond their target domain in exchange for higher-performance [5, 1].However, accelerator deployment in data centers remains limited outside all but the most ubiquitous of application domains like machine learning [5]. High-performance systems require large dice in advanced process nodes with many supporting resources like high-bandwidth memory. As a result, to justify the price that comes with highperformance ic design, designers building non-ml accelerators are left with a choice: either piggyback on ml accelerators-giving up some efficiency but taking advantage of supporting hardware resources-or build more-efficient but less-advanced bespoke hardware.In this work, we show that reconfigurable dataflow accelerators (rdas) are a practical alternative to building fixed-function designs per application domain. We extend Plasticine [5]-a previously proposed ml-focused rda- with low-overhead, microarchitectural extensions to support analytic database queries. These extensions increase area by just 4 %, and the unified accelerator outperforms a multi-core software baseline by 1500 x.Finally, we show how to re-purpose these extensions to implement data structures like trees and hash tables that are critical to asymptotically optimal query plans. We introduce a threading model for vector dataflow accelerators that extracts parallelism from data structures with irregular control flow using fine-grained thread scheduling- outperforming a gpu by 8x.
590 $a School code: 0212.
650 4 $a Flexibility. $3 3560705
650 4 $a Relational data bases. $3 3683439
650 4 $a Databases. $3 747532
650 4 $a Design. $3 518875
650 4 $a Energy efficiency. $3 3555643
650 4 $a Spectrum allocation. $3 3562481
650 4 $a Algorithms. $3 536374
650 4 $a Queries. $3 3564462
650 4 $a Field programmable gate arrays. $3 666370
650 4 $a Car pools. $3 3683472
650 4 $a Information science. $3 554358
650 4 $a Data analysis. $2 bisacsh $3 3515250
650 4 $a Software. $2 gtt. $3 619355
650 4 $a Communication. $3 524709
650 4 $a Bandwidths. $3 3560998
650 4 $a Aggregates. $3 3564159
690 $a 0389
690 $a 0459
690 $a 0723
710 2 $a Stanford University. $3 754827
773 0 $t Dissertations Abstracts International $g 83-03A.
790 $a 0212
791 $a Ph.D.
792 $a 2021
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28688324