CENTER Carnegie Mellon UniversityCarnegie Mellon Computer Science DepartmentSchool of Computer Science
Workshop on Multiserver Scheduling (WORMS)
Related Activities
Outreach Roadshow


Mor Harchol-Balter (CMU, Computer Science), Email:, Phone: (412) 268-7893

Alan Scheller-Wolf (CMU, GSIA), Email:, Phone: (412) 268-5066

Arrangements Coordinator:

Charlotte Yano, Email:, 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 systems
Load balancing and load sharing
Queueing analysis and sensitivity analysis of multiserver systems
Workload characterization for multiserver systems
Computational methods for multi-dimensional optimization
Impact of heavy-tailed workloads on multiserver system performance
Applications of multiserver systems, including:

    Web server farms

    Servers for high-performance computing/grid computing

    Manufacturing applications



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 information.

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: ACTIVITIES





This material is based upon work supported by National Science Foundation under Grant No. 0122581.
Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the
National Science Foundation