[dcchairs2016] UbiComp/ISWC 2016 Doctoral School notification - #112

[dcchairs2016 at ubicomp.org]

Dear Harini Kolamunna,

Please find enclosed the reviews for your submission for the Ubicomp/ISWC 2016 Doctoral School.

112: "Enabling Context-Aware Application Function Virtualization in Wearable Personal Area Networks"

Despite not being able to accept your submission at this year's Ubicomp/ISWC Doctoral School, committee members provided guidance and feedback on your submitted paper. We highly encourage you to follow the valuable advices that the committee member entered in their reviews towards improving on your doctoral work.

Thank you for submitting to the UbiComp 2014 Doctoral School.


Max Mühlhäuser
Nadir Weibel
Rene Mayrhofer

UbiComp 2016 Doctoral School Chairs


------------------------ Submission 112, Review 1 ------------------------

Title: Enabling Context-Aware Application Function Virtualization in Wearable Personal Area Networks


Confidence

   3  (Very confident - I am knowledgeable in the area)

Contribution to UbiComp

   The paper addresses function allocation in personal area networks of
   UbiComp devices. No Hardware issues. OKish for the UbiComp DC from this
   perspective, but not "at the very core".

Overall Rating

   2  (Probably reject: I would argue for rejecting this paper.)

The Review

   The paper addresses PANs (personal area Networks) with "Tier 1 devices"
   (think of Smartphone) that provide PAN-wide allocation of functions to
   devices (while "Tier 2 devices" -think of Smartwatch, Fitband etc. - are
   further candidates for such allocations).

   The idea is good, but the proposed approach is not well presented and -
   what is much more critical - is not convincing at all.

   A formal model for the function placement is presented, basically
   revolving around the different execution costs (called "implementation
   cost") and communication costs for the execution of each function on each
   possible device. Placement is then defined as a MinCost optimization
   problem (function allocation problem) that can be mapped onto a known
   similar problem (uncapacitated facility location). 

   In the context of Ubicomp, it is not sufficient to defined ANY model that
   provides a nice mathematical or algorithmic challenge, it is important to
   define a practically useful model. In this respect, the model presented
   does not address (or is not sufficiently described by the author w.r.t.)
   - resource competition! (several functions executed on same cpu, with
   same remaining battery capacity, or communicated over the same channel,
   etc.)
   - scheduling over time
   - the fact that function calls arrive over time: it is just illusionary
   to think of a system where at time T0 all function requests are known and
   until their completion, no other function requests will arrive; this fact
   seems to be reflected in that iterative execution of the mapping
   algorithm can be performed, but consequences for required re-mapping are
   not properly considered at all
   - Utility functions for conflicting goals
   ... and many more subtle issues known from the Placement and scheduling
   literature.

   Given the highly interesting topic addressed in the PhD research, it is
   suggested to review the model and the general field of research from
   which to "import" approaches, and to resubmit the work. 


------------------------ Submission 112, Review 2 ------------------------

Title: Enabling Context-Aware Application Function Virtualization in Wearable Personal Area Networks


Confidence

   3  (Very confident - I am knowledgeable in the area)

Contribution to UbiComp

   This work is about minimizing energy consumption across a set of PAN
   devices by executing functions such as step detection only on a single
   instead of multiple devices in a PAN. It is clearly relevant to Ubicomp.

Overall Rating

   3  (Maybe reject: I would agree with rejecting this paper.)

The Review

   Although the topic itself is timely and relevant and there is a clear
   technology focus, there are two issues in terms of submitting it to the
   Ubicomp DC:

   a) It is not clear if the motivation for this approach is well aligned
   with what it may be able to deliver. As the main motivation, the author
   uses an example of simple sensors such as accelerometers being present in
   multiple devices in a a PAN, and the hypothesis is that energy can be
   saved by only using one of them. However, that hypothesis is not yet
   proven due to two factors: i) In many current devices, power consumption
   of the accelerometer (for the gyroscope it is different, but I don't
   believe that to carry over nearly as well as the accelerometer) is
   negligible considering the other power consumers. By adding a DSP for
   pre-processing, step detection does not impact battery runtime in any
   significant manner anymore (i.e. not waking the main CPU is the main
   benefit, not turning off the accelerometer itself). ii) The sensors are
   distributed over different body parts, and an accelerometer attached to
   the wrist will be able to give different results than the smartphone in
   the pocket. To be more specific, step detection on a smartphone in a
   lady's handbag is not very effective and less accurate than on a
   wrist-worn watch. That is, the application (even a simple one like step
   counting) may benefit from using multiple sensors to improve its accuracy
   or even to allow some functions.

   b) This is mainly a short version of the paper submitted to the main
   track (which has been conditionally accepted). Therefore, the authors,
   will not benefit significantly from another presentation at the DC, and
   would take the spot of other students who do not get a presentation slot
   in the main program.