Report on
Operating System Support for Planetary-Scale Network Services
In this paper, Bavier et al. describe the functioning and
implementation of PlanetLab. PlanetLab is a group of computers,
each running a modified linux kernel and virtual servers in which
researchers can evaluate and deploy planetary scale services.
Each project is run in a slice of PlanetLab's resources which is meant
to be independent of other slices that may also be using
resources. Additionally, PlanetLab nodes are decentralized from
the main system and provisions must be made for the local management of
the node. The paper expands on these points and describes how
they are implemented in PlanetLab's operating system.
There was a lack of basic information about PlanetLab, such as the
difference between PlanetLab Central (PLC) and nodes running the
PlanetLab operating system. At times, the paper became
exceedingly
detailed in describing such things as the code used to bind a socket to
a port. At other times, there was such a lack of detail that the
granularity used by the scheduler to provide guaranteed CPU time on
nodes was not even mentioned. The paper could have been written
in a much more consistent manner. The evaluation section of the
paper left much to be desired, perhaps it could have benefited from a
case study. In general, the authors could have said more in fewer
pages.
There is a strong case for the usefulness of the paper. Almost
everyone agreed that it would be helpful to users of PlanetLab.
However, the superficial nature and lack of research results called
into question its validity as a research paper. Its usefulness
comes mainly from its technical description of a useful research
platform, not from its own original results. Therefore, perhaps
this paper would be more appropriate as a tech report.
Manfred Georg