BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Firmament: fast\, centralized cluster scheduling at scale - Ionel 
 Gog (Computer Lab)
DTSTART:20161013T140000Z
DTEND:20161013T150000Z
UID:TALK67355@talks.cam.ac.uk
CONTACT:Liang Wang
DESCRIPTION:Scheduling tasks on “warehouse-scale” clusters is a challe
 nging undertaking: thousands of tasks must cleverly and rapidly be placed 
 in order to achieve high utilization and good application-level performanc
 e. Centralized data-center schedulers make high-quality placement decision
 s\, but they come at the cost of high decision latency at scale\, which de
 grades response time for interactive jobs. Distributed schedulers\, by con
 trast\, make low-latency decisions at scale\, but are restricted to simple
  algorithms and can make poor decisions as a result.\n\nIn this talk\, I p
 resent Firmament\, a centralized scheduler that scales to tens of thousand
 s of machines at sub-second latency\, even though it performs a computatio
 nally expensive min cost flow optimization. To achieve this\, Firmament au
 tomatically chooses between different min-cost flow algorithms\, solves th
 e optimization problem incrementally when possible\, and applies problem-s
 pecific optimizations to common min-cost flow algorithms.\n\nOur experimen
 ts with a Google workload trace from a 12\,500-machine cluster show that F
 irmament places tasks in hundreds of milliseconds\, improving decision lat
 ency by more than 10x over Quincy\, a similar prior scheduler. Moreover\, 
 since Firmament can efficiently solve large optimization problems\, it sup
 ports novel scheduler features that would otherwise be too expensive. Firm
 ament outperforms state-of-the-art distributed schedulers in placement qua
 lity without an increase in scheduling decision latency for common industr
 y workloads.
LOCATION:FW26\, Computer Laboratory\, William Gates Building
END:VEVENT
END:VCALENDAR