BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
X-LIC-LOCATION:America/Los_Angeles
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260402T024534Z
LOCATION:3008\, Level 3
DTSTART;TZID=America/Los_Angeles:20250625T114500
DTEND;TZID=America/Los_Angeles:20250625T120000
UID:dac_DAC 2025_sess159_RESEARCH688@linklings.com
SUMMARY:StreamCSD: Host-Transparent SSD Stream Management via In-Storage C
 ontent Learning
DESCRIPTION:Wenjie Li (DapuStor Corporation); Xiang Chen (Huazhong Univers
 ity of Science and Technology); Yelin Shan (DapuStor Corporation); Jiapin 
 Wang (Dapustor Corporation); Yunxin Huang, Yafei Yang, and Tao Lu (DapuSto
 r Corporation); and You Zhou and Fei Wu (Huazhong University of Science an
 d Technology)\n\nWrite amplification (WA) from migrating valid pages durin
 g garbage collection (GC) degrades SSD performance and lifespan. Although 
 stream management based on high-level software semantics reduces WA, exist
 ing solutions require host modifications, hindering their adoption. We int
 roduce StreamCSD, an SSD-autonomous stream management approach using in-st
 orage content learning, eliminating host-side changes. Leveraging compress
 ion ratios from embedded compressors in computational storage drives (CSDs
 ), StreamCSD employs a streaming K-means algorithm to cost-efficiently clu
 ster data into streams. Evaluations show that StreamCSD reduces WA from 1.
 7 to 1.06 under multimodal generative AI workloads, matching state-of-the-
 art methods with minimal impact on bandwidth. StreamCSD operates without h
 ost modifications, promoting broader adoption of multi-stream SSDs.\n\nTop
 ics: Systems\n\nTracks: SYS5: Embedded Memory and Storage Systems\n\nSessi
 on Chairs: Nima TaheriNejad (Heidelberg University) and Jalil Boukhobza (E
 NSTA)\n\n
END:VEVENT
END:VCALENDAR
