*Looking over the evolution of Internet workloads*
Virgilio Almeida
Federal University of Minas Gerais, Brazil
SWS Colloquium
Virgílio Almeida is a professor of computer science at the Federal
University of Minas Gerais, Brazil. He has held
visiting-professor positions at Boston University and Polytechnic
University of Catalunya, Barcelona, as well as
visiting appointments at Xerox PARC and Hewlett-Packard Research
Laboratory and Polytechnic Institute of NYU. His
research interests include models to analyze the behavior of large-scale
distributed systems. Almeida is a
recipient of a Fulbright Research Scholar Award and is a full member of
the Brazilian Academy of Sciences. He is
also an International Fellow of the Santa Fe Institute for 2008/2009 and a
member of the editorial board of
Internet Computing and First Monday. Almeida is the author of more than
100 technical papers and co-author (with
Danny Menasce) of four books, including Performance By Design and Capacity
Planning for Web Services: Metrics,
Models, and Methods, published by Prentice Hall and translated into three
languages.
The Internet has a number of popular applications and services
experiencing workloads with very different,
non-trivial and unique properties. The emergence of new applications and
computing models (e.g., online social
networking, streaming video, games and cloud computing), and the explosive
growth in popularity of others (e.g.,
search, peer-to-peer, e-business, malware), most of which with workloads
with not fully understood fundamental
properties, make this a research topic of timely relevance. Real workload
characterization and modeling provide
key insights into the cost-effective design, operation, and management of
Internet-based services and systems.
This talk looks over the evolution of Internet workloads, by presenting an
overview of a variety of real Internet
workloads and how they have evolved through the years; from system
workload to social workload. It shows the main
properties of these workloads and discusses the invariants across
different types of workloads. It outlines
methodologies and techniques used in workload characterization and
modeling. Constructing a model involves
tradeoffs between usefulness and accuracy. The talk shows how
characterization techniques have been used to
capture the most relevant aspects of Internet workloads while keeping the
model as simple as possible. The talk
concludes showing some examples of how workload models have been used to
design efficient Web systems and
services.