[Colloquium] Zhang/MS Presentation/Aug 6, 2018

[Margaret Jaffey via Colloquium]

This is an announcement of Chaojie (Sam) Zhang's MS Presentation.

------------------------------------------------------------------------------
Date:  Monday, August 6, 2018

Time:  3:00 PM

Place:  Ryerson 255

M.S. Candidate:  Chaojie Zhang

M.S. Paper Title: MANAGING THE VALUE OF VOLATILE CLOUD RESOURCES:
INFORMATION DISCLOSURE AND GUARANTEE-PRESERVING MANAGEMENT

Abstract:
Cloud providers sell unreliable or “volatile” resources that are
unused by foreground (reserved/high priority) workloads. The value
users can extract from these resources depends on the (i) volatile
resource management algorithm, and (ii) information provided to users
about the resources as statistical guarantees. We describe and
evaluate four volatile resource management approaches (Random, FIFO,
LIFO, LIFO-pools) using commercial cloud resource traces from 608
Amazon EC2 instance pools, for a 3 month period from 5/2017 to 8/2017
from Amazon’s four US regions, each of which contains 2-6 availability
zones. We also consider the value of several information models (MTTR,
limited statistics, Full distribution, and Oracle) that statistically
characterize the resources. Our results show volatile resource
management algorithms can increase user value by 30 to 45% in four
instance exemplars. For example, LIFO and FIFO vs Random can make more
than 2x difference. Slightly richer information models (90pctile)
combined with LIFO and LIFO-pools volatile resource management
increase user value by as much as 10-fold. Our results suggest that
cloud providers should pay significant attention to what statistical
information they provide to users. Then our results of relative user
value per resource-hour for the four exemplars show that 90pctile
information model is best in all cases and achieves close to the max
possible. Hence, simple statstics , such as 90th percentile, can
increase achievable value by 10% to as much as 5x. However, skewed
distribution can lead to misleading information, and thus sharply
reducing derived value. And, these results broadly characterize the
vast majority (475 of 608) of instance pools. The results are the same
ordering for VRMs and information models as in two exemplars, and the
frequency of various relations that are key conclusions for the
exemplars are carried along in the vast majority of instance pools.
Futhermore, we provide a detailed drill-down showing how the volatile
resource management algorithms affect resource interval durations, and
thus potential user value. We further show how the information model
shapes user targeting, success rate, and user value. Finally, we study
offline and two online algorithms to maintain statistical guarantees
in the face of foreground load changes. Our offline algorithm results
show that it is feasible to fully preserve statistical guarantees
under foreground load changes by delaying the release of each volatile
resources to users by a short period of time with trivial resource
waste, from 0.5% to 8.9% in extreme cases, at the same time increasing
user value by up to 134%. Moreover, two online algorithms, AIMD
algorithm and Distribution Targeting algorithm, can dynamically
preserve guarantees with modest resource waste, and in doing so
increase user value by up to 82%. This suggests that further research
exploring online algorithms is promising.

Chaojie's advisor is Prof. Andrew Chien

Login to the Computer Science Department website for details:
 https://www.cs.uchicago.edu/phd/ms_announcements#chaojie

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Margaret P. Jaffey            margaret at cs.uchicago.edu
Department of Computer Science
Student Support Rep (Ry 156)               (773) 702-6011
The University of Chicago      http://www.cs.uchicago.edu
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=