USENIX Association
Proceedings of MobiSys 2003:
The First International Conference on
Mobile Systems, Applications, and Services
San Francisco, CA, USA
May 5-8, 2003
' 2003 by The USENIX Association All Rights Reserved For more information about the USENIX Association:
Phone: 1 510 528 8649 FAX: 1 510 548 5738 Email: office@usenix.org WWW: http://www.usenix.org
Rights to individual papers remain with the author or the author’s employer.
Permission is granted for noncommercial reproduction of the work for educational or research purposes.
This copyright notice must be included in the reproduced paper. USENIX acknowledges all trademarks herein.

MobiSys 2003: The First International Conference on Mobile Systems, Applications, and Services USENIX Association
273
Tactics-Based Remote Execution for Mobile Computing
Rajesh Krishna Balan
†
, Mahadev Satyanarayanan
†‡
, SoYoung Park
†
, Tadashi Okoshi
†
†
Carnegie Mellon University and
‡
Intel Research Pittsburgh

rajesh,satya,seraphin,slash
@cs.cmu.edu
Abstract
Remote execution can transform the puniest mobile device
into a computing giant able to run resource-intensive applica-
tions such as natural language translation, speech recognition,
face recognition, and augmented reality. However, easily par-
titioning these applications for remote execution while retain-
ing application-specific information has proven to be a difficult
challenge. In this paper, we show that automated dynamic re-
partitioning of mobile applications can be reconciled with the
need to exploit application-specific knowledge. We show that
the useful knowledge about an application relevant to remote
execution can be captured in a compact declarative form called
tactics. Tactics capture the full range of meaningful partitions
of an application and are very small relative to code size. We
present the design of a tactics-based remote execution system,
Chroma, that performs comparably to a runtime system that
makes perfect partitioning decisions. Furthermore, we show
that Chroma can automatically use extra resources in an over-
provisioned environment to improve application performance.
1 Introduction
Remote execution can transform the puniest mobile de-
vice into a computing giant. This would enable resource-
intensive applications such as natural language trans-
lation, speech recognition, face recognition, and aug-
mented reality to be run on tiny handheld, wearable
or body-implanted platforms. Nearby compute servers,
connected through a low-latency wireless LAN, can pro-
vide the CPU cycles, memory, and energy needed for
such applications.
Unfortunately, two annoying facts cloud this rosy fu-
ture. First, the optimal partitioning of an interactive
application into local and remote components is highly
application-specific and platform-specific. Since mo-
bile hardware evolves rapidly, this optimal partitioning
changes on the timescale of months rather than years.
Suboptimal partitioning can result in sluggish and intol-
erable interactive response. Hence, a tight and ongoing
coupling between application developers and hardware
platform developers appears inevitable. Second, matters
are made worse by the fact that mobile environments ex-
hibit highly variable resource availability. Bandwidth,
energy and presence of compute servers can change on
the timescale of minutes or hours, as a user moves to
different locations. Re-partitioning an application for
changed operating conditions at this timescale is there-
fore essential. These considerations suggest that an au-
tomated approach to partitioning applications for remote
execution is necessary. However, partitioning an appli-
cation without taking into consideration its unique char-
acteristics may result in sub-optimal partitions.
Can automated dynamic re-partitioning be reconciled
with the need to exploit application-specific knowledge?
In this paper, we show that this is indeed possible. Our
key insight is that the knowledge about an application
relevant to remote execution can be captured in compact
declarative form that is very small relative to code size.
More specifically, the full range of meaningful partitions
of an application can be described in a compact external
description called remote execution tactics or just tactics
for brevity. Thus, the tactics for an application consti-
tute the limited and controlled exposure of application-
specific knowledge necessary for making effective parti-
tioning and placement decisions for that application in a
mobile computing environment.
In this paper, we examine three applications of the
genre mentioned earlier (natural language translation,
speech recognition, and face recognition) and show that
the tactics for each is much less than one percent of total
code size. We present the design of Chroma, a tactics-
based remote execution system, and show that sound par-
titioning and placement of these applications using tac-
tics is possible. We show that Chroma is able to achieve
application performance that is comparable to execution
on an ideal runtime system.
In addition, we show that Chroma can opportunisti-
cally utilize extra resources in an over-provisioned envi-
ronment. This allows us to achieve lower latencies for
the three applications mentioned above.
The rest of this paper is organized as follows: Sec-
tion 2 presents the assumptions and goals of this work
while Section 3 presents the design of Chroma. We
present our experimental setup in Section 4. Sections 5
and 6 present Chroma’s performance relative to an ideal
runtime system. In Section 7, we show how tactics can
improve application performance in the presence of extra
resources. Section 8 presents related work and Section 9
concludes the paper.