Transcript Aggregating Network Performance in Distributed

A Portal Based Approach to
Viewing Aggregated Network
Performance Data in Distributed
Brokering Systems
By
Gurhan Gunduz, Shrideep Pallickara, Geoffrey Fox
Syracuse University,
Indiana University, Community Grid Labs
ggunduz, spallick, [email protected]
IC 2003
LAS VEGAS, NV, USA
Introduction
More applications, services and
frameworks becoming network centric.
Network performance is important.
Efforts on network measurement
IP provider Metrics(subgroup of IETF’s
Bench Marking Working Group)
CADIA(Cooperative Association for
Internet Analysis Data)
NWS(Network Weather System)
NaradaBrokering
Distributed event brokering system designed to
run on a large network of cooperating broker
nodes.
Broker nodes are organized in a cluster-based
architecture which allows the system to support
large heterogeneous client configurations.
Communication in NaradaBrokering is
asynchronous.
NaradaBrokering provides support for JMS, P2P
interactions, audio-video conferencing while
supporting communication through firewalls and
proxies
NaradaBrokering Transport
Framework
Transport framework aims to abstract the
operations that need to be supported for
enabling efficient communications between
nodes. TCP, UDP, SSL, RTP and HTTP.
Operations that need to be supported between
two communication endpoints are encapsulated
within the “link” primitive.
A Link is an abstraction that hides details
pertaining to communications between 2
communicating entities.
Link and Performance
Measurement
Can expose and measure a set of performance
factors.
Cooperation from the other end-point of the
communication link.
Echo behavior
Can measure round trip delays, jitter, bandwidth,
loss rates, etc.
Links can enable/disable the measurement of all
performance factors or a specific factor.
Accumulating performance metrics
for a node
Every broker incorporates a Monitoring
Service(MS).
Transport Controller of a node maintains
the list of the links
MS cycles through the links and retrieve
performance information.
MS controls frequency of metric
measurements.
Aggregating performance metrics
from multiple nodes
MS report performance data to a
Performance Aggregation Service(PAS).
PAS exchanges information with MS.
PAS can give simple commands to MS.
PASs can exchange information with each
other
Encapsulating performance data
MS encapsulates performance data in an
XML format.
Why XML;



Easy access to relevant fields in the
performance data.
Description capability of the content provides
intelligent data mining.
XPATH queries
PAS
Broker
Node
Broker
Node
Monitoring
Service
Link
Data
Link
Data
Aggregates info
from nodes in a
certain domain
Performance Aggregation
Service
Control Message
Exchange
Multiple PAS
Broker
Node
Monitoring
Service
Control Message
Exchange
PAS
PAS
Aggregates info
from nodes in a
certain domain
PAS
info in
terch
ange
PAS
PAS
Aggregating performance metrics
from multiple nodes
Aggregated data is saved in a database
Currently flat file
We plan using light weight XML database
for this purpose.


Apache Xindice
SourceForge exist
Accumulation of data in a portal
Information accumulated within
Aggregators is accessible from a portal
Apache Jetspeed is used as a portal
environment.
Portal view
Accumulation of data in a portal II
Portals can display multiple HTML
Can collect content from disparate remote
sources.
Can facilitate customized user groups.
Restrictive user view using customized view
capability of a portal
Portlets, a specialized module, Java servlet
which operates in a portal, is used to view
accumulated performance data within the
aggregators.
Accumulation of data in a portal III
XSLT portlet is used to view aggregated
performance data.
XSLT portlet converts a given XML file into
HTML using the given XSL style sheet.
Accumulation of data in a portal IV
Two ways for customized user view


Having separate database file for each user
group
Let portlet select it from one database file
Detecting conditions
Evaluate constraints in the aggregation node.
Adv?
Use XPATH to query our database.
Check metrics for thresholds
Inform nodes to take actions to correct situation
Future Work
Trade-offs of using flat files versus lightweight databases.
Identify, circumvent, project and prevent
system bottlenecks.
Aid routing algorithms.