Full metadata record

DC Field Value Language
dc.contributor.author Kim, Minsub ko
dc.contributor.author Lee, Sungjin ko
dc.date.accessioned 2021-01-29T07:23:23Z -
dc.date.available 2021-01-29T07:23:23Z -
dc.date.created 2020-10-15 -
dc.date.issued 2020-08-25 -
dc.identifier.citation 11th ACM SIGOPS Asia-Pacific Workshop on Systems, APSys 2020, pp.90 - 97 -
dc.identifier.isbn 9781450380690 -
dc.identifier.uri http://hdl.handle.net/20.500.11750/12876 -
dc.description.abstract Most recommender systems are designed to comply with service level agreement (SLA) because prompt response to users' requests is the most important factor that decides the quality of service. Existing recommender systems, however, seriously suffer from long tail latency when the embedding tables cannot be entirely loaded in the main memory. In this paper, we propose a new SSD architecture called EMB-SSD, which mitigates the tail latency problem of recommender systems by leveraging in-storage processing. By offloading the data-intensive parts of the recommendation algorithm into an SSD, EMB-SSD not only reduces the data traffic between the host and the SSD, but also lowers software overheads caused by deep I/O stacks. Results show that EMB-SSD exhibits 47% and 25% shorter 99th percentile latency and average latency, respectively, over existing systems. © 2020 ACM. -
dc.language English -
dc.publisher Association for Computing Machinery -
dc.title Reducing Tail Latency of DNN-based Recommender Systems using In-storage Processing -
dc.type Conference -
dc.identifier.doi 10.1145/3409963.3410501 -
dc.identifier.scopusid 2-s2.0-85092208085 -
dc.type.local Article(Overseas) -
dc.type.rims CONF -
dc.description.journalClass 1 -
dc.contributor.localauthor Lee, Sungjin -
dc.identifier.citationStartPage 90 -
dc.identifier.citationEndPage 97 -
dc.identifier.citationTitle 11th ACM SIGOPS Asia-Pacific Workshop on Systems, APSys 2020 -
dc.identifier.conferencecountry JA -
dc.identifier.conferencelocation Tsukuba -

qrcode mendeley

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.