BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/New_York BEGIN:DAYLIGHT DTSTART:20220313T030000 TZOFFSETFROM:-0500 TZOFFSETTO:-0400 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZNAME:EDT END:DAYLIGHT BEGIN:STANDARD DTSTART:20221106T010000 TZOFFSETFROM:-0400 TZOFFSETTO:-0500 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZNAME:EST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20230629T001435Z UID:D0CAF510-D3E3-4836-83F1-1CE00751D569 DTSTART;TZID=America/New_York:20220427T183000 DTEND;TZID=America/New_York:20220427T200000 DESCRIPTION:Abstract:\n\nSolid-state drives (SSDs) are everywhere! Flash-ba sed SSDs have established themselves as a higher-performance alternative t o hard disk drives in cloud and mobile environments. SSDs are widely used as a form of storage in mobile devices\, laptops\, digital cameras\, and c loud servers. Hence\, improving the performance of SSDs impacts the overal l computing system and the experience of millions of end-users. SSDs deliv er significantly higher speeds and are more reliable than HDDs\, however\, they still remain a performance bottleneck of computing systems. SSDs are relatively reliable\; however\, they still fail\, which can result in dat a loss or system unavailability. Datacenter operators are interested in pr edicting future drive failures to administer drive replacement\, data migr ation\, and drive acquisition strategies. The talk describes my research a ddressing the challenges of improving the reliability and response time of flash-based storage systems using machine learning.\n\nTo improve reliabi lity\, we propose a machine learning based approach for automatically pred icting SSD failures. We analyzed telemetric data collected from over 30\,0 00 drives running live applications in data centers over a span of six yea rs\, to find the most critical reasons for SSD failures. We introduce an a pproach for automatically predicting future SSD failures in data centers w hich enable interpretability of the model's predictions. To improve respon se time\, we propose a neural network based approach to improve prefetchin g in SSDs. Prefetching is a technique to speed-up fetch operations by pred icting future block accesses and preloading them into the main memory ahea d of time. This research identifies the challenges of prefetching in SSDs and explains why prior approaches fail to achieve high accuracy and presen ts a deep neural network (DNN) based prefetching approach that significant ly outperforms the state-of-the-art. I will conclude my talk with research challenges that I plan to address in the future.\n\nSpeaker(s): Chandrani l “Nil” Chakraborttii\, \n\nVirtual: https://events.vtools.ieee.org/m/ 311981 LOCATION:Virtual: https://events.vtools.ieee.org/m/311981 ORGANIZER:dr.a.rusu@ieee.org SEQUENCE:2 SUMMARY:Improving response time\, reliability\, and lifetime of flash drive s using machine learning URL;VALUE=URI:https://events.vtools.ieee.org/m/311981 X-ALT-DESC:Description: <br /><p class="yiv7120620870p1"><strong>Abstract:< /strong></p>\n<p class="yiv7120620870p1">Solid-state drives (SSDs) are eve rywhere! Flash-based SSDs have established themselves as a higher-performa nce alternative to hard disk drives in cloud and mobile environments. SSDs are widely used as a form of storage in mobile devices\, laptops\, digita l cameras\, and cloud servers. Hence\, improving the performance of SSDs i mpacts the overall computing system and the experience of millions of end- users. SSDs deliver significantly higher speeds and are more reliable than HDDs\, however\, they still remain a performance bottleneck of computing systems. SSDs are relatively reliable\; however\, they still fail\, which can result in data loss or system unavailability. Datacenter operators are interested in predicting future drive failures to administer drive replac ement\, data migration\, and drive acquisition strategies. The talk descri bes my research addressing the challenges of improving the reliability and response time of flash-based storage systems using machine learning.</p>\ n<p class="yiv7120620870p2">&nbsp\;</p>\n<p class="yiv7120620870p1">To imp rove reliability\, we propose a machine learning based approach for automa tically predicting SSD failures. We analyzed telemetric data collected fro m over 30\,000 drives running live applications in data centers over a spa n of six years\, to find the most critical reasons for SSD failures. We in troduce an approach for automatically predicting future SSD failures in da ta centers which enable interpretability of the model's predictions. To im prove response time\, we propose a neural network based approach to improv e prefetching in SSDs. Prefetching is a technique to speed-up fetch operat ions by predicting future block accesses and preloading them into the main memory ahead of time. This research identifies the challenges of prefetch ing in SSDs and explains why prior approaches fail to achieve high accurac y and presents a deep neural network (DNN) based prefetching approach that significantly outperforms the state-of-the-art. I will conclude my talk w ith research challenges that I plan to address in the future.</p> END:VEVENT END:VCALENDAR