{"619405":{"#nid":"619405","#data":{"type":"event","title":"PhD Defense by Mohan Kumar","body":[{"value":"\u003Cp\u003ETitle: Taming Latency In Data Center Applications\u003Cbr \/\u003E\r\n\u003Cbr \/\u003E\r\nMohan Kumar Kumar\u003Cbr \/\u003E\r\nSchool of Computer Science\u003Cbr \/\u003E\r\nCollege of Computing\u003Cbr \/\u003E\r\nGeorgia Institute of Technology\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E------------------------------\u003Cbr \/\u003E\r\n\u003Cbr \/\u003E\r\nDate: Monday, April 1st, 2019\u003Cbr \/\u003E\r\nTime: 10:00 AM - Noon EDT\u003Cbr \/\u003E\r\nLocation: KACB 2100\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E------------------------------\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cbr \/\u003E\r\nCommittee:\u003Cbr \/\u003E\r\n---------------\u003Cbr \/\u003E\r\nDr. Taesoo Kim (Advisor, School of Computer Science, Georgia Tech)\u003Cbr \/\u003E\r\nDr. Ada Gavrilovska (School of Computer Science, Georgia Tech)\u003Cbr \/\u003E\r\nDr. Umakishore Ramachandran (School of Computer Science, Georgia Tech)\u003Cbr \/\u003E\r\nDr. Tushar Krishna (School of Electrical Engineering, Georgia Tech)\u003Cbr \/\u003E\r\nDr. Keon Jang (Max Planck Institute for Software Systems, Saarbr\u0026uuml;cken)\u003Cbr \/\u003E\r\n\u003Cbr \/\u003E\r\nAbstract:\u003Cbr \/\u003E\r\n------------\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003EA new breed of low-latency I\/O devices, such as the emerging remote memory access\u003Cbr \/\u003E\r\nand the high-speed Ethernet NICs, are becoming ubiquitous in current data centers. For\u003Cbr \/\u003E\r\nexample, big data center operators such as Amazon, Facebook, Google, and Microsoft are\u003Cbr \/\u003E\r\nalready migrating their networks to 100G. However, the overhead incurred by the system\u003Cbr \/\u003E\r\nsoftware, such as protocol stack and synchronous operations, is dominant with these faster I\/O devices. This dissertation approaches the above problem by redesigning a protocol stack to provide an interface for the latency-sensitive operation, and redesigning synchronous operation such as TLB shootdown and consensus in the operating systems and distributed systems respectively.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cbr \/\u003E\r\nFirst, the dissertation presents an extensible protocol stack, Xps to address the software\u003Cbr \/\u003E\r\noverhead incurred in protocol stacks such as TCP and UDP. Xps provides the abstractions\u003Cbr \/\u003E\r\nto allow an application-defined, latency-sensitive operation to run immediately after the\u003Cbr \/\u003E\r\nprotocol processing (called the fast path) in various protocol stacks: in a commodity OS\u003Cbr \/\u003E\r\nprotocol stack (e.g., Linux), a user space protocol stack (e.g., mTCP), as well as recent smartNICs. For all other operations, Xps retains the popular, well-understood socket interface. Xps\u0026rsquo; approach is practical: rather than proposing a new OS or removing the socket interface completely, our goal is to provide stack extensions for latency-sensitive operations and use the existing socket layer for all other operations. Second, the dissertation provides a lazy, asynchronous mechanism to address the system\u003Cbr \/\u003E\r\nsoftware overhead incurred due to a synchronous operationTLB shootdown. The key idea\u003Cbr \/\u003E\r\nof the lazy shootdown mechanism, called Latr, is to use lazy memory reclamation and\u003Cbr \/\u003E\r\nlazy page table unmap to perform an asynchronous TLB shootdown. By handling TLB\u003Cbr \/\u003E\r\nshootdowns in a lazy fashion, Latr can eliminate the performance overheads associated with IPI mechanisms as well as the waiting time for acknowledgments from remote cores. By proposing an asynchronous mechanism, Latr provides an eventually consistent solution to TLB shootdowns. Finally, the dissertation untangles the logically coupled consensus mechanism from the application which alleviates the overhead incurred by consensus algorithms such as Multi-Paxos\/Viewstamp Replication(VR). By physical isolation, Dyad eliminates the consensus component from competing for system resources with the application which improves the application performance. To provide physical isolation, Dyad defines the abstraction needed from the SmartNIC and the operations performed on the application running on the host processor. With the resulting consensus mechanism, the host processor handles only the client requests on the host processor in the normal case and the disappropriate messages needed for consensus is handled on the SmartNIC.\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n","summary":null,"format":"limited_html"}],"field_subtitle":"","field_summary":"","field_summary_sentence":[{"value":"Taming Latency In Data Center Applications"}],"uid":"27707","created_gmt":"2019-03-19 14:18:37","changed_gmt":"2019-03-19 14:18:37","author":"Tatianna Richardson","boilerplate_text":"","field_publication":"","field_article_url":"","field_event_time":{"event_time_start":"2019-04-01T11:00:00-04:00","event_time_end":"2019-04-01T13:00:00-04:00","event_time_end_last":"2019-04-01T13:00:00-04:00","gmt_time_start":"2019-04-01 15:00:00","gmt_time_end":"2019-04-01 17:00:00","gmt_time_end_last":"2019-04-01 17:00:00","rrule":null,"timezone":"America\/New_York"},"extras":[],"groups":[{"id":"221981","name":"Graduate Studies"}],"categories":[],"keywords":[{"id":"100811","name":"Phd Defense"}],"core_research_areas":[],"news_room_topics":[],"event_categories":[{"id":"1788","name":"Other\/Miscellaneous"}],"invited_audience":[{"id":"78761","name":"Faculty\/Staff"},{"id":"78771","name":"Public"},{"id":"174045","name":"Graduate students"},{"id":"78751","name":"Undergraduate students"}],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[],"email":[],"slides":[],"orientation":[],"userdata":""}}}