Middleware for QoS Adaptive Server Design
|
in collaboration with
|
|
Internet is rapidly becoming a medium for conducting business transactions
and selling priced services. When financial considerations are involved
a need arises for providing consumer satisfaction guarantees commensurate
with service price and customer importance.
Quality of Service (QoS) becomes an
important consideration. Hosting current and
emerging Internet services is a
growing area that will need to consider QoS guarantees and provide
appropriate tools for monitoring and
enforcement of QoS contracts between the service host
and the service outsourcing party. Today's Internet servers are not
designed for QoS guarantees. Support for QoS is generally
absent from both the end-system and
Internet infrastructure. While, in the long term, market forces may
create enough momentum to provide and use this support, in the immediate
future a lot of the QoS management functionality can be achieved in
a middleware layer that lies transparently
between the existing service and the OS. The middleware approach can be used
to provide transparent load monitoring,
overload protection, dynamic quality adaptation, QoS isolation, and service
differentian in web servers without modifying server code, underlying
protocol, or the OS scheduling and communication resource management
mechanisms. As such, the approach can be immediately deployed in the
present-day Internet with no changes to the client or server.
This work is conducted in collaboration with Hewlett-Packard
Laboratories.
Students: Matthew Nahlik
Publications:
-
Tarek F. Abdelzaher and Chenyang Lu,
``Modeling and Performance Control of Internet Servers,''
Invited Paper, 39th IEEE Conference on
Decision and Control, Sydney, Australia, December 2000.
-
Tarek F. Abdelzaher,
``An Automoated Profiling Subsystem for QoS-Aware Services,''
IEEE Real-Time Technology and
Applications Symposium, Washington D.C., June 2000.
-
Tarek F. Abdelzaher, Nina Bhatti,
``Adaptive Content Delivery for Web Server QoS,''
International Workshop on Quality of Service,
London, UK, June 1999.
-
Tarek F. Abdelzaher, Nina Bhatti,
``Web Content Adaptation to Improve Server Overload Behavior,''
International World Wide Web Conference,
Toronto, Canada, May 1999.
-
Tarek F. Abdelzaher, Kang G. Shin,
``QoS Provisioning with qContracts in Web and
Multimedia Servers,''
IEEE Real-Time Systems Symposium, Phoenix, Arizona,
December 1999.
Robust Server Farms
Present-day servers (such as web and multimedia servers)
fail ungracefully under overload conditions
as the number of clients exceeds server capacity causing undesirable
service outage. In performance critical applications, such as e-commerce
and guaranteed web hosting, poor overload behavior is unacceptable.
I am investigating robust server architectures where overload behavior
is predictable and robust performance guarantees are achieved. Issues
that complicate building robust servers include dynamic content generation
or distillation, load balancing, and interaction with back-end components
such as databases which can become performance bottlenecks. A solution to
the robust server problem should address issues of QoS-provisioning, QoS
adaptation, and load balancing across multiple servers as well as between
front-end and back-end servers.
Operating System Support for Server QoS
The recent years have seen several interesting efforts on OS support for
server platforms. Servers (e.g., web or e-commerce servers) exercise
their platform resources in a manner very different from clients. While
many OS mechanisms have been developed to deal with performance requirements
on the client side, more research is needed on server-side operating systems.
The topic deals with identifying the demands and performance
characteristics of server-side software and providing the necessary
operating system mechanisms to meet server-side performance requirements.
Publications:
-
Tarek F. Abdelzaher, Kang G. Shin,
``End-host Architecture for QoS-Adaptive Communication,''
IEEE Real-Time Technology and
Applications Symposium, Denver, Colorado, June 1998.
-
A. Mehra, A. Shaikh, T. Abdelzaher, Z. Wang, and K. Shin,
``Realizing Services for Guaranteed-QoS
Communication on a Microkernel Operating System,'' IEEE Real-Time
Systems Symposium, Madrid, Spain, December 1998.
-
Tarek Abdelzaher, Anees Shaikh, Farnam Jahanian, and Kang Shin,
``RTCast: Lightweight Multicast for Real-Time Process Groups,''
IEEE Real-Time Technology and Applications
Symposium, Boston, MA, June 10-12, 1996.
Winner of the 1996 RTAS Best Student Paper Award
(Earlier
version appears in Tech Report CSE-TR-291-96, Department of Electrical
Engineering and Computer Science, University of Michigan, March, 1996.
Real-Time Scheduling Theory
Real-time scheduling theory is perhaps one of the most studied and
well understood aspects of real-time computing. Most scheduling problems
are NP-complete, which typically calls for heuristic solutions. We
demonstrate that computationally tractable ``optimal'' solutions to
the general problem of distributed resource constrained communicating task
allocation and scheduling are also possible if the search space is
appropriately defined and the search algorithm has certain desirable
properties. This is reported in the papers below.
-
Tarek F. Abdelzaher,
``A Schedulable Utilization Bound for Aperiodic Tasks,''
University of Virginia, Technical Report CS-2000-21,
August 2000.
-
Tarek F. Abdelzaher and Kang G. Shin,
``Period-Based Partitioning and
Assignment for Large Real-Time Applications,''
IEEE Transactions on Computers, Vol. 49, No. 1,
January 2000.
- Tarek F. Abdelzaher, Kang G. Shin,
"Combined Task and Message Scheduling
in Distributed Real-Time Systems," IEEE
Transactions on Parallel and Distributed Systems,
November, 1999.
- D.T. Peng, K.G. Shin, T.F. Abdelzaher,
"Assignment and Scheduling of Communicating Periodic Tasks in
Distributed Real-Time Systems"
IEEE Transactions on Parallel and Distributed
Systems, Vol. 8, No. 12, December 1997.
- Tarek F. Abdelzaher, Kang G. Shin,
"Comment on 'A Pre-Run-Time Scheduling Algorithm for Hard Real-Time
Systems,'"
IEEE Transactions on Software Engineering, Vol. 23, No. 9,
September, 1997.
- Tarek F. Abdelzaher, A.M. Salem, S.A. Hammad and M.A. Sheirah
"A Priority-Based Scheduler for Dynamic Real-Time Tasks with Simple
Resourse Requirements" in Proc 3rd IASTED International
Conference, Cairo, Egypt, December, 1994.
QoS-Sensitive Web-Caching
With the dramatic increase of web traffic on the Internet, proper
web caching schemes become crucial to providing adequate service performance
to the client. Presently, web caches treat all client requests
alike without regard to client importance and available resources.
In a future where Internet services are priced and differentiated
service support is in place, caching policies should adapt not only to
clients' access patterns but also to the importance or QoS class of
clients. I would like to investigate the problem of providing
statistical performance guarantees to different classes of clients
via appropriate QoS-sensitive caching policies.
Students: Christopher Sullivan, Vinesh Rathore
QoS-Portable Services
A question that has faced the developers of performance-critical services
for a long time is how to develop commercial guaranteed-QoS software on
commercial-off-the-shelf hardware such that it provides performance
guarantees on arbitrary target platforms. The answer lies partly in
efficient mechanisms for adapting QoS guarantees to
arbitrary platform resource types, capacity, and load. Collectively,
these mechanisms will ensure QoS portability, i.e., portability
of the software's performance in addition to the portability of
its functionality. QoS portability is a
new software engineering concern motivated by
the new multimedia and guaranteed-QoS applications that need to be
designed in a cost-efficient manner to run in a QoS-guaranteed fashion
on platforms whose speed and resources are not known at software
design time.
Publications:
