Latency
Profiles: Performance Monitoring for Wide Area
Applications
Recent technological advances
have enabled the deployment of wide area
applications against Internet accessible sources.
A performance challenge to such applications is
unpredictable behavior, e.g., the end-to-end
latency of accessing sources in the dynamic WAN.
Our objective is to exploit open protocols and to
use passive and non-intrusive information
gathering mechanisms to learn end-to-end latency
distributions and construct Latency Profiles
(LPs) corresponding to these distributions. We
hypothesize that groups of clients
(client-cluster), within an autonomous system
(AS), that are accessing a content server, in
another AS, may be represented by a single LP.
Related networking research on IDMaps, points of
congestion, and BGP routes support such a
hypothesis.
For more information: http://www.umiacs.umd.edu/research/CLIP/Handle/
Profile-Based
Data Delivery in Wide Area and Mobile Environments
An important characteristic of data access in both
wide area and mobile environments is that clients
often have different preferences for the latency
and recency of their data. Traditional
caching and scheduling schemes do not provide
interfaces for clients to express these
preferences, and treat all client and applications
alike. In this research we develop a data
delivery scheme using client profiles, a set of
parameters that allow clients to express
preferences for their different applications.
A cache uses profiles to determine whether to
deliver a cached object to the client or to
download a fresh object from a remote server.
We develop an architecture for profiles that is
both scalable and straightforward to implement at
a cache. We also study how profiles can be
used to support diverse mobile applications. We
develop a scheme that uses profiles for both
caching and scheduling data delivery on the
wireless bandwidth.
For more information: http://www.cs.umd.edu/~bright/research.html
Resource
Profiles: Monitoring Resources for Improved
Performance
Resource profiles consist of
actively gathered information regarding temporal
patterns of availability and updates of resources.
With such a profile at hand, mediators can
optimize query performance using estimated
resource availability. The main challenge we face
is the immense effort that is needed to maintain
resource profiles of some tens of millions of
resources. Therefore, brute force solution is
replaced with an intelligent algorithm that bases
future resource probing on past performance. This
work is complementary to the work on profile-based
data delivery (http://www.cs.umd.edu/~bright/research.html).
Resource profiles serve as the basis for choosing
among various policies for pulling servers.
Efficient
Algorithms for Computing Resource Availability and
Update Patterns
With the
immense volume of resources available on the
Internet, frequently updated, changed, and
eventually expired, efficient and effective
information delivery becomes a true challenge.
This task is shared by many prevalent applications
such as Web crawlers and Web caches. Mediators
were proposed in supporting access to objects and
provide integration among data sources. We propose
to extend mediator's responsibility beyond its
typical services, to improve information delivery
by estimating resource availability and update
patterns. Typically, such services require
background information on the server's performance
(to measure latency) and availability. The aim of
this project is to propose efficient and
configurable algorithms for background collection
of such knowledge on servers' patterns of behavior
(such as regular periodic updates, or abrupt
changes), to improve the accuracy and efficiency
of information delivery. The verification of some
tens of millions of resources and their current
values cannot be handled by brute force and
requires the use of resource prioritization. For
more information contact Avigdor
Gal.
|