04/26/2000 - Research paper reviews

Bryan Wang (bryan@clip.dhs.org)
Thu, 27 Apr 2000 00:11:26 -0700

<<grapevine.txt>> <<emerald.txt>>
Bryan Wang, 4/26/00

Experience with Grapevine: The Growth of a Distributed System
----------------------------------------------------------------------

Grapevine is an interesting system because it is designed to grow and
evolve, with careful consideration to resource management. The goal
was to allow the number of users to grow arbitrarily high and to
support the growth by adding servers with fix resources (not more
powerful servers).

Resource management turned out to be trickier than the developers
thought, and they made several policy changes on the system after the
system grew to a certain size. Due to the nature of the system, many
problems appeared only as the system grew beyond a certain capacity,
and were not easily predictable under "lab" conditions.

The paper was well written and very readable. Its ideas both on how
to and how not to design distributed systems were equally educational.
In the cases where the system didn't provide perfect performance, the
authors did a good job explaining the situation and rationalized the
design tradeoffs involved.

An interesting point the authors bring up in the concluding remarks is
that they are "slowly forgetting the details of implementation". I've
seen and experienced this phenomonon in my professional endeavors, and
can truely sympathize with legacy code lasting forever!

Bryan Wang, 4/26/00

Fine-Grained Mobility in the Emerald System
----------------------------------------------------------------------

Emerald is an attempt to improve performance of mobile processing
through object-granular mobility. Objects are large or small, and the
minimum amount of data is transmitted (only object-related data is
transmitted, not code). The underlying design constraint is to avoid
degradation of processes running on a local node, and offload the
burden of mobility on only those objects that require it.

The designers tried to enhance the performance of process migration by
providing fine-grained mibility. The additional benefits of
fine-grained mobility are data movement, invocation performance, and
garbage collection. They were successful in acheiving these goals.

This paper is filled with good ideas on mobile processing yet I found
it to be poorly organized. It suffered from some of the same problems
as the StarOS paper (intermixing high and low level concepts).