In 1998 and 1999, research within Stieltjes on heavy tails in queues has focused on two topics. The first topic is, whether other service disciplines than FCFS may lead to a less detrimental response time behaviour. For the queue with the Processor Sharing discipline  and the queue with the Last-Come-First-Served Preemptive Resume discipline , important disciplines that both play a key role in queueing networks of so-called product-form, the response time tail behaviour turns out to be regularly varying of index iff the service time tail behaviour is regularly varying of index . Hence these disciplines result in better tail behaviour of sojourn times than FCFS. In a series of studies (see, e.g., ) we have also investigated the workload tail behaviour of service systems with Generalized Processor Sharing (GPS). GPS-based scheduling algorithms, such as Weighted-Fair-Queueing, have emerged as an important mechanism for achieving differentiated Quality-of-Service in networks with integrated traffic services. We have identified conditions under which, in GPS, traffic of one class does not suffer from heavier-tailed traffic of another class.
The second topic
is the heavy-traffic behaviour
of a queue with heavy-tailed service time distribution.
A queueing system is said to be in heavy traffic when its traffic load
This topic is of theoretical interest, since
in the traditional heavy-traffic limit theorems
it is assumed that the second moments of service and
interarrival times are finite, whereas (2)
with leads to an infinite second moment.
The topic is also of practical interest,
since heavy-traffic limit theorems may give rise to
useful approximations  in situations with a reasonably
light traffic load.
One can identify a coefficient of contraction
times the waiting time
has a proper limiting distribution for
One of the results obtained in  reads:
For the stable FCFS queue with regularly varying service time
with mean and
of index , as specified in (2),
the `contracted' waiting time
. The limiting distribution is
specified by its Laplace-Stieltjes transform
and the coefficient of contraction is the root
of the equation
with , with the property that
This result strongly differs from the case of a service time
distribution with finite variance; there the coefficient of contraction
behaves like instead of
and the limiting distribution is the negative exponential distribution
(hence with Laplace-Stieltjes distribution ).
Many challenging unsolved problems remain. In particular, one would like to know the tail behaviour of the workload in the case of fluid queues fed by multiple sources, some of which have heavy-tailed on-periods; and the related problem of the asymptotics of multiserver queues with heavy-tailed service time distributions also remains open.