Today’s businesses are forming complex multi-partner commercial
arrangements which involve some level of integration of their computing
systems; at the same time governments are building “systems of
systems” to integrate legacy applications with new capabilities. We
see an increasing trend towards the aggregation of disparate systems,
and their applications. The result is the complex integration of
independently developed computing systems to create new capabilities; we
call this Aggregation of Systems.
One effect of aggregation is that network traffic becomes
unpredictable as application data patterns interfere with one another;
applications that ran successfully on a LAN experience failures in an
aggregated system. A second effect is that Quality of Service parameters
are lost when a request transits from one infrastructure to another. In
many cases it is impossible to determine if the system will be able to
meet end-to-end QoS needs. This is especially serious for applications
that have end-to-end timeliness requirements.
In other words, we do not know how to build these systems so that we
know they will work! That is the reason for the Aggregate Systems
Challenge and this Requirements Gathering Session for that Challenge.
Introduction
Dock Allen, Mitre Corporation, introduced the session.
Her PowerPoint slides are here.
The aim of the exercise is to demonstrate support for an end-to-end
QoS across heterogeneous systems.
There are several definitions of 'Real Time', and the first three
are the key aspects to be considered
- Time critical - the usefulness of data and operations degrades
over time
- Temporal responsiveness, where the other side will 'time out' if
response is too slow
- Accurate pacing of data streams - such as streaming video
- Time sensitive - data goes stale
- Temporal coherence - eg image fusion
- Workflow based on temporal relationships
Timeliness of data/operations may be handled differently in each
enclave, so how do we build real-tome systems that span enclaves?
Available technologies include
- Splice
- Real Time Corba
- RT Java
- RT Message Oriented Middleware
- RT extensions to UML
- Transport options including IPv6 with priority support,
DiffServ, MPLS, U4EA, InfiniBand
Also some interesting reseach is being done in some organizations,
such as DARPA QORUM/QOIN adaptive resource management, and the current
MITRE research on dynamic scheduling.
The plan is to issue a challenge, supported by the OMG, to
prototype and demonstrate partial solutions, with the aim of
identitying gaps
This needs to include
- Multiple enclaves
- Heterogeneous infrastructures
- Dependable end-to-end timeliness required for some of the
applications
- Performance challenging but achievable
- Involves enterprise integration
Goals
Dave Lounsbury introduced the goal of the challenge - to demonstrate
practical approaches to guaranteeing end-to-end timeliness in
applications running in complex aggregated environments. PDF
slides are here.
The goal of the session is to identify the requirements of the
problem space, to select or synthesize a small number of use cases,
and to drive a statement of requirements based on the experience of
users.
The characteristics of the environment are
- predictable response critical
- multiple programming and communications paradigms e.g. database,
remote invocations, message passing
- multi vendor
- distributed
Dave illustrated the sort of system he had in mind by describing a
radar system on a ship which which is being attacked by, for instance,
a missile. This is a multi-stage problem, with a number of components
that ware necessary for a solution to work.
Another example in the defense world is FCS (Future Combat
Systems), implemented in battle tanks at present, which could be used
in more diverse units such as C&C platforms, indirect fire
vehicles, troops mounted and on the ground, mobile robotic sensors,
and ground sensors. This scenario has a number of different
dimensions, and is further complicated by the fact that it operates in
a hostile environment in which there may be jamming of electronic
signals.
There are also commercial implications, and Dave illustrated the
case of an aerospace manufacturer in which the effective working of
the factory floor depended on the effectiveness of a wide range of
other systems. There is a flow of a transactions through a range
of systems each of which has to respond fast enough to keep
construction working.
Next steps
- Brainstorming on example problem
- Requirements gathering
- Objectives
- Format
- Guidelines
- Identification of resources
- Planning
Requirements
Dock Allen then presented the requirements for a solution - her
PowerPoint slides are here:
End-to-end honoring of QoS
- Passing QoS parameters between nodes and between software environments
- Translation of QoS parameters into the local representation for
each node or environment
Top-to-bottom Honoring of QoS
- Passing QoS parameters up and down between infrastructure layers
- Translation of QoS parameters into layer-specific
representations
- The ideal in that all layers should honor QoS, but this is often
not the case
Local honoring of QoS requires, for instance,
- Preempting the current tasks for one with a higher QoS
- QoS sensitive queue management
- QoS-based resolution of resource contention
- Ability to set QoS of internal threads
There is also a requirement to support more than one policy and to
resolve conflicts between them.
Network Management has a role to pay. Sometimes, network
attempts to handle local overloads can exacerbate them; on the other
hand, networks may be able to even out loads.
Joe Bergmann described the work he had been doing with many
universities in the US, UK and Japan, with the aim of having an event in
Austin. He had also spoken with HP (Walter Stahlecker) and IBM (MaryAnne
Fisher). He was intending to speak with the US Air Force, which
has no activities in this space, and expected the active involvement of
the Army, which was already working.
Sally Long pointed out that the scenarios described so far were huge,
and that in order to have a real challenge the scope needed to be
reduced considerably.
The intension is to have a workshop in Austin, but some work needs to
be done between now and then.
Notes from a Brainstorming Session
- University of Idaho, Access GRID Notes
- Austin Workshop
- Pare down Application to subsystems for challenge
- Specific subsystems with particular class of QoS problem
- Need few vendors in the beginning
- Value prop for vendors demonstrate product works
- End-to-end QoS guarantee is main focus; through at least 2-3
subsystems
- How to allocate resources across heterogeneous environments to
guarantee QoS
- Determine Application Problem
- Find sponsors / coordinator before Austin
- Distributed Heterogeneous Resource Management (Managing network
& resources, applications
- NASA-RM interest adaptive resource management
- Sensor web wrt satellite
- NDP (Network Transport over UDP)
- Prioritization scheme
- Reusable scope necessary to limite resource investment
- Airforce research - future global grid, network-centric adaptive
resource management
- Security Constraints effect resource management
- NSWC - ship radar tracking dynamically, ie switching from plane to
ship tracking dynamically recreates data paths
Several people offered their help: Jim Alves Foss, Frank Boyle,
Dock Allen, Dave Marlow, Michael Mimette, Gavin Watt, Loni Welsh (Joe
Bergmann will confirm).
There was a discussion about whether any of the organizations
represented in the room were working on a problem which could provide a
scenario for the Challenge. Anyone able to help was asked to mail
the group at qos_realtime@opengroup.org.
