Building Robust Network Services Through Efficient Resource Management (Thesis)

Abstract:

Network services have been increasingly integrated into our daily
lives, but their accessibility and stability are also frequently
impacted by flash crowds or Denial of Service (DoS) attacks. To be
immune from flash crowds or DoS attacks, robust network services must
possess two important qualities: completeness and
generality. Completeness implies that all resources must be protected,
including CPU time, memory and disk capacity, and link bandwidth.
Generality means not handling attacks or faults as extraordinary
events, but instead treating them within the same framework used
during normal operations. Fundamentally, this is a matter of efficient
management of networked resources.

Toward this end, we use Content Distribution Networks (CDN) as an
example, and investigate how request redirection impacts CDN
robustness. CDN systems deploy redundant resources (servers)
geographically distributed across the Internet and distribute client
requests to an appropriate server based on a variety of
factors---e.g., server load, network proximity, cache locality---in an
effort to reduce response time and increase the system capacity under
load. We explore the design space of the redirection strategies
employed by request redirectors, and define a class of new algorithms
that carefully balance load, locality, and proximity. We use
large-scale detailed simulations to evaluate various strategies. These
simulations demonstrate the effectiveness of our new
algorithms, which yield a 60-91% improvement in system capacity when
compared with the best published CDN technology, yet user-perceived
response latency remains low and the system scales well with the
number of servers. We also build a prototype CDN, named CoDeeN, on
the PlanetLab testbed. CoDeeN helps us to gain experience on managing
and monitoring an operational CDN, and will be used in future research.

Through this thesis, we demonstrate that the resilience of large wide
area network services can be improved through efficient management of
networked resources. By adapting unified resource management schemes,
we present a practical way to build network systems that not only
handle a larger volume of regular traffic more easily, but also absorb
flash crowds and deter DoS attacks as a natural part of their
operations.