BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Building Latency-Critical Datacenter Systems
DTSTART:20210112T140000
DTEND:20210112T150000
DTSTAMP:20260501T095117Z
UID:08deaf1827ccf04401bc2c46e42d19a90a802e981fcb13a4bd34ae37
CATEGORIES:Conferences - Seminars
DESCRIPTION:Marios Kogias\, graduated PhD from EDIC\nAbstract\nOnline serv
 ices play a major role in our everyday life for communication\, entertainm
 ent\, socializing\, e-commerce\, etc.  These services run inside the data
 center under strict tail-latency service level objectives in order to rema
 in interactive. The emergence of new hardware for IO has enabled microsec
 ond-scale datacenter communications that challenge the efficiency of exist
 ing operating system and network mechanisms. Also\, new in-network progra
 mmable devices start being deployed in datacenters and introduce a new com
 puting paradigm that shifts functionality traditionally performed at the e
 nd-points to the network.\nIn this talk will I revisit the operating syste
 ms\, networking\, and\, distributed systems infrastructure specifically ta
 rgeting latency-critical datacenter systems\, while drawing intuition from
  basic queueing theory results. In the first part of the talk\, I will fo
 cus on ZygOS[SOSP 2017]\, a system optimized for μs-scale\, in-memory com
 puting on multicore servers. ZygOS implements a work-conserving scheduler
  within a specialized operating system designed for high request rates and
  a large number of network connections. ZygOS revealed the challenges ass
 ociated with serving remote procedure calls (RPCs) on top of a byte-stream
  oriented protocol\, such as TCP. In the second part of the talk\, I will
  present R2P2[ATC 2019]. R2P2 is a transport protocol specifically design
 ed for datacenter RPCs\, that exposes the RPC abstraction to the endpoints
  and the network\, making RPCs first-class datacenter citizens. R2P2 enab
 les pushing functionality\, such as scheduling\, fault-tolerance\, and tai
 l-tolerance\, inside the transport protocol\, making it application-agnost
 ic. I will show how using R2P2 allowed us to offload RPC scheduling to pr
 ogrammable switches that can schedule requests directly on individual core
 s.\n\n\n 
LOCATION:zoom
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR
