BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/New_York BEGIN:DAYLIGHT DTSTART:20230312T030000 TZOFFSETFROM:-0500 TZOFFSETTO:-0400 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZNAME:EDT END:DAYLIGHT BEGIN:STANDARD DTSTART:20231105T010000 TZOFFSETFROM:-0400 TZOFFSETTO:-0500 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZNAME:EST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20231121T223024Z UID:0F714493-D883-4C6C-876A-605AF89F8D84 DTSTART;TZID=America/New_York:20231031T090000 DTEND;TZID=America/New_York:20231031T094500 DESCRIPTION:The unprecedented surge in the scale of network topologies and traffic has fueled a pressing demand for rapid simulation of new network d esigns\, prior to both emulation and real-world deployment. However\, conv entional network simulation techniques do not scale well\, due to their hi gh computational costs and low degrees of parallelism. In response\, netwo rk performance estimators have emerged as a promising alternative\, levera ging parallel inference of deep neural networks for acceleration. However\ , these estimators still face challenges in achieving satisfactory scalabi lity and generality.\n\nIn this talk\, we introduce some of our recent wor k on improving the performance\, scalability and generality of network per formance estimation. I will first introduce DeepQueueNet\, our work on usi ng deep neural network models to emulate the behaviour of network devices\ , supporting arbitrary network topologies and device traffic management me chanisms. For generating training data samples\, the design and implementa tion of DeepQueueNet depended upon our recent work on a new Python-based d iscrete event simulator\, which I developed from scratch. Our extensive ex periments showed that DeepQueueNet achieved near-linear speedup in the num ber of GPUs\, and its estimation accuracy for average and 99th percentile round-trip time outperforms existing end-to-end DNN-based performance esti mators. I will conclude the talk with key insights in our ongoing work to improve DeepQueueNet.\n\nSpeaker(s): Baochun Li\n\nNiagara Falls\, Ontario \, Canada LOCATION:Niagara Falls\, Ontario\, Canada ORGANIZER:rboutaba@uwaterloo.ca SEQUENCE:14 SUMMARY:Towards Fast\, Scalable and Generalized Network Performance Estimat ion URL;VALUE=URI:https://events.vtools.ieee.org/m/385214 X-ALT-DESC:Description: <br /><p>The unprecedented surge in the scale of ne twork topologies and traffic has fueled a pressing demand for rapid simula tion of new network designs\, prior to both emulation and real-world deplo yment. However\, conventional network simulation techniques do not scale w ell\, due to their high computational costs and low degrees of parallelism . In response\, network performance estimators have emerged as a promising alternative\, leveraging parallel inference of deep neural networks for a cceleration. However\, these estimators still face challenges in achieving satisfactory scalability and generality.</p>\n<p>In this talk\, we introd uce some of our recent work on improving the performance\, scalability and generality of network performance estimation. I will first introduce Deep QueueNet\, our work on using deep neural network models to emulate the beh aviour of network devices\, supporting arbitrary network topologies and de vice traffic management mechanisms. For generating training data samples\, the design and implementation of DeepQueueNet depended upon our recent wo rk on a new Python-based discrete event simulator\, which I developed from scratch. Our extensive experiments showed that DeepQueueNet achieved near -linear speedup in the number of GPUs\, and its estimation accuracy for av erage and 99th percentile round-trip time outperforms existing end-to-end DNN-based performance estimators. I will conclude the talk with key insigh ts in our ongoing work to improve DeepQueueNet.</p> END:VEVENT END:VCALENDAR