CloneCloud Discussion Summary

[Zikai]

We discussed CloneCloud together with MAUI, which are two similar
systems to offload mobile applications to computation cloud to achieve
performance and energy gains. In this summary, I will focus on
differences between the two systems and potential problems with
CloneCloud paper.

The difference between MAUI and CloneCloud mainly lies in: (1) MAUI
focus on energy saving while CloneCloud focus on improving latency.
MAUI’s performance gain is a byproduct of its energy saving policies
while CloneCloud authors assert that their method can also be applied
to save energy. (2) To run mobile applications on the cloud, MAUI
requires them to be written in Microsoft .NET Common Language Runtime
(CLR) while CloneCloud depends on migration of application VMs to run
on device VMs. (3) MAUI has both offline and online procedures while
CloneCloud is purely offline. (4) MAUI needs manual annotations to
mobile applications while CloneCloud does not.

The first potential problem with CloneCloud is that authors do not
mention how to migrate a multi-thread mobile application. Though most
mobile applications on smart phones are single threaded, more and more
multi-threaded mobile applications show up with development of smart
phone hardware. For a multi-threaded mobile application, the problem
is when multiple threads of execution depend on some shared state and
some of the threads are migrated while some are not, how to maintain
correctness of the application.

The second potential problem is when CloneCloud generates the profile
trees with its Dynamic Profiler, authors do not mention coverage of
their input data. This is the same problem as test coverage. If
coverage is low, the system may not bring much performance gain. If
coverage is high , the partition database may be too large to fit in
smart phone’s precious storage. How to do this with the maximum
efficiency is important.