The rise of cloud computing has enabled governments and companies to store huge
data sets. Today, one of the major performance bottlenecks processing these
data sets is the garbage collector of managed runtime environments such as the
Java virtual machine. The garbage collector must now scan heaps of tens of
gigabytes of memory, which causes intolerable pauses to the user. A recent
study that was conducted jointly with the University of Neuchatel showed that
database NoSQL Cassandra could introduce pauses of up to 6 minutes with a heap
of a hundred gigabytes.

The aim of this internship is to propose and study a new garbage collection
algorithm that can support large memory loads. The algorithm starts from the
principle that the garbage collector can scan objects concurrently with the
application if these objects are not accessed by the application at the same
time. The aim of the internship will be to find an algorithm to separate the
heap into two parts, one with the objects accessed by the application , the
other with non-accessed objects.

Developments will be made in the HotSpot Java Virtual Machine (JVM) in C++. It
is therefore requested the applicant to have an excellent level of programming
and knowledge in memory collection.