(CMU, Computer Science), Email: email@example.com,
Phone: (412) 268-7893
Scheller-Wolf (CMU, GSIA),
Email: firstname.lastname@example.org, Phone: (412) 268-5066
Charlotte Yano, Email: email@example.com, Phone: (412) 268-7656,
FAX (412) 268-5576
The study of queues with multiple servers dates back over fifty
years, to the seminal work of Kiefer and Wolfowitz. While their
work was appreciated at the time, it is doubtful that anyone could
have foreseen the vast and elemental role these systems have taken
in society, including everything from call centers to checkout
lines to Web server farms to high-performance supercomputing systems.
The popularity of this type of architecture is understandable,
as it allows for increased performance, while being cost-effective
and easily scalable. This is particularly true in computer systems,
such as Web server farms.
Given the ubiquity of multiserver systems, it is surprising
that even at this late date so little is understood regarding
their performance as compared with their single-server counterpart.
While for complex single server systems, we have closed-form solutions
for busy periods, response times (sojourn times), and much more,
we cannot make similar statements for most two-server systems.
Furthermore while the M/G/1 queue is easily analyzed under a range
of scheduling and prioritization policies, for the M/G/2 queue
most scheduling policies aren't even close to analytically tractable.
The intractability of multiserver systems under different
scheduling policies is particularly unfortunate, since "smart"
scheduling algorithms have proven extremely effective at reducing
mean response time for single-server queues, and are likely to
be even more effective in the multiserver regime, due to the added
flexibility multiple servers offer. There is a growing body of
evidence suggesting that choosing the right task assignment policy
when assigning jobs to hosts in a server farm can improve mean
response time by orders of magnitude. Yet many important task
assignment policies are still analytically intractable. Even the
simplest questions regarding multiserver systems, such as ''how
does one powerful server compare with many less-powerful servers''
and ''what is the benefit of allowing cycle stealing between the
servers'' remain largely open.
However recent advances in the area of multiserver systems
(including moment results, tail asymptotic results for some specialized
systems, methods for reducing 2-dimensionally infinite Markov
chains to 1-dimensionally infinite Markov chains, time and unit-scaling
techniques, and new results in scheduling theory and task assignment)
lead us to believe that we are now finally in a position to solve
some of these central multiserver questions. The purpose of this
workshop is to bring together select researchers from around the
globe to present their research and exchange thoughts, ideas,
conjectures, and analytical techniques related to multiserver
systems and their applications. We believe that significant progress
on these difficult problems will be made through the exchange
of ideas, and the collaborations borne out from this workshop.
Performance modeling techniques for server farms
Scheduling/Prioritization algorithms for multiserver
Load balancing and load sharing
Queueing analysis and sensitivity analysis of
Workload characterization for multiserver systems
Computational methods for multi-dimensional optimization
Impact of heavy-tailed workloads on multiserver
Applications of multiserver systems, including:
The list of workshop talk abstracts is shown here: TALKS
. The following confirmed speakers are shown here: SPEAKERS
. The format/schedule for the workshop is shown here: PROGRAM
. Although our speaking slots are now full, we naturally
welcome other researchers in the field of multiserver scheduling
who may want to attend the workshop. Please register below and contact
the chairs for possible partial financial assistance. A list of
all people registered is provided here: attendees
The workshop will consist of 1/2-hour invited talks over a two-day
period with plenty of time for discussion. Airfare and hotel for
all invitees will be arranged for and paid for by CMU.
If you plan to attend the workshop, you must REGISTER HERE . Laptop projectors, overhead
projectors, and a white board will be available. Speakers with
electronic presentations should email these to us by APRIL 5.
Please click on EMAILING PRESENTATIONS for more
For directions from the Pittsburgh Airport to the Holiday
Inn Hotel click here: DIRECTIONS
TO HOTEL. Please save your transportation receipts and give
these to Charlotte Yano for reimbursement.
For current weather in Pittsburgh: WEATHER
For a list of popular attractions in the Pittsburgh area: