Database Reference
In-Depth Information
In the case of an in-line visualization annotation, one could consider deploy-
ing a host of visualization-related functions—changing color depth, changing
frame rate, changing resolution, visualization-specific compression techniques,
and so forth. Based on the user-specified priorities (color is unimportant, but
frame rate is crucial), the in-transit manipulation of the extracted data al-
lows for a much higher fidelity interaction for the application scientist. As
the adaptation of the data stream becomes more complex, it leads naturally
to discussion of full-fledged autonomic control of the network and computer
platforms, which is the topic of the next sections.
5.2.4.3
Autonomic Data Movement Services Using IQ-Paths
Among data-driven high-performance applications, such as data mining and
remote visualization, the ability to provide quality of service (QoS) guarantees
is a common characteristic. However, due to most networking infrastructure
being a shared resource, there is a need for middleware to assist end-user
applications in best utilizing available network resources.
An IQ-Path is a novel mechanism that enhances and complements existing
adaptive data streaming techniques. First, IQ-Paths dynamically measure
25
,
26
and also predict the available bandwidth profiles on the network links. Sec-
ond, they extend such online monitoring and prediction to the multilink paths
in the overlay networks used by modern applications and middleware. Third,
they offer automated methods for moving data trac across overlay paths,
including splitting a data stream across multiple paths and dynamically dif-
ferentiating the volume and type of data trac on each path. Finally IQ-
Paths use statistical methods to capture the noisy nature of available network
bandwidth, allowing a better mapping to the underlying best-effort network
infrastructure.
The overlay implemented by IQ-Paths has multiple layers of abstraction.
First, its
middleware underlay
—a middleware extension of the network un-
derlay proposed in Reference 27—implements the execution layer for overlay
services. The underlay is comprised of processes running on the machines
available to IQ-paths, connected by logical links and/or via intermediate pro-
cesses acting as router nodes. Second, underlay nodes continually assess the
qualities of their logical links as well as the available resources of the machines
on which they reside. Figure 5.4 illustrates an overlay node part of an IQ-Path.
The routing and scheduling of application data is performed with consider-
ation of path information generated by the monitoring entities. The service
guarantees provided to applications are based on such dynamic resource mea-
surements, runtime admission control, resource mapping, and a self-regulating
packet routing and scheduling algorithm. This algorithm, termed PGOS (pre-
dictive guarantee overlay scheduling), provides probabilistic guarantees for the
available bandwidth, packet loss rate, and round-trip time (RTT) attainable
across the best-effort network links in the underlay. More information on IQ-
Paths can be found in Reference 28.
