BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Latency-driven Performance in Data Centres - Diana Popescu\, Compu
 ter Lab
DTSTART:20200604T140000Z
DTEND:20200604T150000Z
UID:TALK143965@talks.cam.ac.uk
CONTACT:Srinivasan Keshav
DESCRIPTION:Data centre-based cloud computing has revolutionised the way b
 usinesses use computing infrastructure. Instead of building their own data
  centres\, companies rent computing resources and deploy their application
 s on cloud hardware. Providing customers with well-defined application per
 formance guarantees is of paramount importance to ensure transparency and 
 to build a lasting collaboration between users and cloud operators. A user
 ’s application performance is subject to the constraints of the resource
 s it has been allocated and to the impact of the network conditions in the
  data centre. \nFirstly\, I show how to use the Precision Time Protocol (P
 TP)\, through an open-source software implementation PTPd\, to measure net
 work conditions. I propose PTPmesh\, which uses PTPd\, as a cloud network 
 monitoring tool for tenants. Furthermore\, I conduct a measurement study u
 sing PTPmesh in different cloud providers\, finding that network latency v
 ariability in data centres is still common. Normal latency values in data 
 centres are in the order of tens or hundreds of microseconds\, while unexp
 ected events\, such as network congestion or packet loss\, can lead to lat
 ency spikes in the order of milliseconds. Secondly\, I show that network l
 atency matters for certain distributed applications even in small amounts 
 of tens or hundreds of microseconds\, significantly reducing their perform
 ance. Given the network latency variability observed in data centres\, app
 lications’ performance is determined by their placement within the data 
 centre. Thirdly\, I propose latency-driven\, application performance-aware
 \, cluster scheduling as a way to provide performance guarantees to applic
 ations. I introduce NoMora\, a cluster scheduling architecture that levera
 ges the predictions of application performance dependent upon network late
 ncy combined with dynamic network latency measurements taken between pairs
  of hosts in data centres to place applications\, showing that NoMora impr
 oves application performance by choosing better placements than other sche
 duling policies.\n\nBio: \n\nDiana Andreea Popescu is a postdoctoral resea
 rcher working on IoT\, edge computing and data centre networking in the Sy
 stems Research Group at the University of Cambridge\, Department of Comput
 er Science and Technology. She completed her PhD in Computer Science with 
 a dissertation titled "Latency-driven performance in data centres" in the 
 same group. She studied for her undergraduate degree in Computer Science a
 t the University Politehnica of Bucharest\, Romania\, and for her master d
 egree in Computer Science at École Polytechnique Fédérale de Lausanne\,
  Switzerland. She interned several times at Google in the Network Infrastr
 ucture team and at CERN during her studies. Her research interests are dat
 a centre networking\, network measurement\, IoT and edge computing.
LOCATION:Webex Meeting number: 140 388 203 Password: ZApmfTb7h56 https://m
 eetingsemea10.webex.com/meetingsemea10/j.php?MTID=mf0c350ff1fab96540218f57
 67c51d098
END:VEVENT
END:VCALENDAR