Next: Bird's
eye view and Up:
Introduction Previous: Introduction
Design of high level QoS goals to low level storage actions transformation mechanisms in management software is done by experts with many years of prior experience as system architects and administrators. However, even the experts are encountering the following types of problems while designing robust storage management systems:
Complexity: The level of details,
required to write the specifications is non-trivial. ECA rules are written as a
certain storage management action being taken when a system observable violates
a predetermined threshold. The transformation code is specified as ECA rules (e.g.
if 
AND 
, then increase data
prefetching size by 20%) where actions are taken upon the violation of
threshold values. It is difficult for the composers of ECA rules to: (1) choose
which combination of system parameters to observe from a large set of possible
observables; (2) determine appropriate threshold values after considering the
interaction of a large set of system variables; and (3) select a specific
corrective action from the large set of competing options. As the number of
users, storage devices, storage management actions, and service level
agreements increase, it becomes computationally exhaustive for system
administrators or storage management tool developers to consider all the
alternatives.
Brittleness: It is difficult for vendors to provide prepackaged transformation code with their products because this code becomes brittle with respect to changing system configurations, user workloads, and department/business constraints. It is difficult for the storage management vendors to envision all of the potential use case scenarios ahead of time, and thus, many of the current storage management solutions provide work-flow environments which, in turn, pass the responsibility of transforming high level QoS goals (via work-flow scripts) to an organization's system administrators and infrastructure planners. To summarize, we restate the discussion-panel conclusion in FAST'03: Existing storage management frameworks are like a ``four-year old kid''- ``They mess up more than they are actually useful.''
Next: Bird's eye view and Up:
Introduction Previous:
Introduction