BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:A selective and biased choice of techniques for building a distrib
 uted data store - Paweł T. Wojciechowski\, Poznan University of Technolog
 y
DTSTART:20240514T110000Z
DTEND:20240514T120000Z
UID:TALK216898@talks.cam.ac.uk
CONTACT:Jamie Vicary
DESCRIPTION:Single-machine data stores cannot support the scale and ubiqui
 ty of data\ntoday. The Internet applications and services must process a h
 uge number\nof concurrent requests and events per second. So\, they use di
 stributed (or\nreplicated) data stores which store and process data on mul
 tiple machines\,\noffering key advantages in performance\, scalability\, a
 nd reliability. The\npurpose of the talk is to present a selective and bia
 sed choice of\ntechniques and results which can be used for building an ef
 ficient\ndistributed data store. Biased\, because I only present solutions
  and\nresults developed within a research project that I did with my PhD\n
 students. Selective\, because an exhaustive description would be too\nexha
 usting to fit into a single talk. Therefore I will be discussing just\nthe
  design of our novel database index for key-value data store systems\,\nan
 d only skim our other contributions that are directly related to\ndistribu
 ted systems. The index\, called Jiffy\, has been designed with\nperformanc
 e and scalability in mind. Therefore it has been designed as a\nlock-free 
 concurrent data structure\, which can dynamically adapt to the\nchanging w
 orkload. It achieves superior performance despite built-in\natomic operati
 ons (batch updates\, snapshots\, and range scans). During the\ntalk I will
  be presenting Jiffy's architecture\, the algorithms for\ninserting and lo
 oking up the key-value pairs\, and the operations used for\nresizing the d
 ata structure dynamically. The other contributions of our\nproject include
 : efficient support for replica state recovery after\nfailures\, either by
  extending the classic Paxos consensus algorithm\, or\nthrough the use of 
 persistent memory\, and a bit surprising theoretical\nresults which are ap
 plicable to distributed data store systems that\ncompromise consistency in
  favour of high availability and speed\, but also\nsupport operations ensu
 ring strong consistency (which requires consensus\namong replicas). (Based
  on a keynote talk at DEBS '23)
LOCATION:SS03\, Computer Laboratory
END:VEVENT
END:VCALENDAR