BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Bayesian Inference in Networks of Queues - Dr Charles Sutton (Edin
 burgh)
DTSTART:20100309T110000Z
DTEND:20100309T120000Z
UID:TALK22558@talks.cam.ac.uk
CONTACT:Zoubin Ghahramani
DESCRIPTION:Diagnosis of performance problems in computer systems is a ric
 h application area for machine learning\, because data about system perfor
 mance can be readily obtained. Many such diagnostic questions concern syst
 em performance in the face of load\, for example\, diagnosing performance 
 bottlenecks during past workload spikes\, or diagnosing bottlenecks during
  particularly slow requests. Diagnostic problems are also of practical int
 erest\, because much of the cost of managing a system is due to finding an
 d recovering from failures.\n\nModel-based approaches are attractive here\
 , because they can incorporate human knowledge about the system\, and do n
 ot require labeled failure data. A classical family of models of computer 
 performance is queueing models. Queueing models predict the explosion in s
 ystem latency under high workload in a way that is often reasonable for re
 al systems.\n\nIn this talk\, we present a novel graphical modeling viewpo
 int on queueing models\, which allows them to be used for inference about 
 past system behavior and learning from incomplete data. The idea is to mea
 sure a small set of arrival and departure times from the system\, treating
  the times that were not measured as missing data. The posterior distribut
 ion over missing data and parameters can then be sampled using Markov chai
 n Monte Carlo techniques. Developing a sampler in this case is significant
 ly more challenging than for standard graphical models\, because of the co
 mplex deterministic dependencies that arise in queueing models. On data fr
 om a benchmark Web 2.0 application\, we demonstrate the ability to localiz
 e performance problems with 25% of the measurement overhead.\n
LOCATION:Engineering Department\, CBL Room 438
END:VEVENT
END:VCALENDAR