|
SMART: Storage Management Analytics and Reasoning Technology
Capacity planning, application/storage performance management,
backup/restore operations, configuration management, security and
availability analysis are some of the key storage management responsibilities
of a system administrator. Typically, storage administrators write scripts
that automate many of these storage management tasks. As the number of
business service level agreements, department policies, quality of
service (QoS) goals, storage devices, protocols, applications and users
increases, it becomes difficult for system administrators to ensure
performance, provisioning, availability and security goals by using
ad-hoc script writing approaches management scenarios.
SMART is a self-evolving, corrective action engine that optimizes
storage-resource allocation in a fully automated, cost-efficient way so
most clients experience predictable performance in their accesses to a
shared, large-scale storage utility. Hardware costs play a dwindling
role relative to managing costs in current enterprise systems. Static
provisioning approaches are far from optimal, given the high burstiness
of I/O workloads and the inadequate available knowledge about storage
device capabilities. Furthermore, efficient static allocations do not
contemplate hardware failures, load surges and workload variations;
system administrators must currently deal with those by hand, as part
of a slow and error-prone observe-analyze-act loop. Prevalent access
protocols (e.g.,SCSI and Fibre Channel) and resource-scheduling
policies are largely best-effort. Unregulated competition is
unlikely to result in a fair, predictable resource allocation. Previous
work on this problem includes management policies encoded as sets of
rules, heuristics-based scheduling of individual I/Os, decisions based
purely on feedback loops and on the predictions of models for system
components. The resulting solutions may: not be adaptive at all (as in
the case of rules), or be dependent on models that are costly to
develop, or be ignorant of the system's performance characteristics, as
observed during its lifetime.
Our ongoing research effort on SMART encompasses several areas:
- Applying machine-learning techniques for creating self-evolving
mathematical models for storage components, workloads and actions.
- Developing formulations for exploring possible solutions for
deciding corrective actions and their invocation parameters.
- Applying constraint optimization and planning algorithms to
decide the optimal answer.
- Evaluating the working of SMART algorithms on a setup with
real-world storage systems.
If you are working on similar areas and interested in collaborating
with us, please send an email to Sandeep Uttamchandani (sandeepu@us.ibm.com)
IBM Almaden Research - Storage Management and Solutions
|
Sandeep Uttamchandani, Li Yin, Guillermo Alvarez, John Palmer, Gul Agha.
"Chameleon: a self-evolving, fully-adaptive resource arbitrator for storage systems,"
USENIX Technical Conference, Anaheim, CA, April , 2005.
Sandeep Uttamchandani,, Kaladhar Voruganti, Sudarshan M. Srinivasan, John Palmer, David Pease.
"Polus: Growing Storage QoS Management beyond a "4-year Old Kid","
3rd USENIX Conference on File and Storage Technologies (FAST '04) , 2004.
Li Yin, Sandeep Uttamchandani, John Palmer, Randy Katz, Gul Agha.
"AutoLoop: Automated Action Selection in the Observe-Analyze-Act Loop of Storage Systems,"
6th IEEE International Workshop on Policies for Distributed Systems and Networks (POLICY'05), April 2005.
Sandeep Uttamchandani, Xiaoxin Yin, John Palmer, Gul Agha.
"MonitorMining: Creating Domain Knowledge for System Automation Using a Gray-box Approach,"
In proceedings of the Ninth IFIP/IEEE International Symposium on Integrated Network Management, Nice, France, May 2005.
Lin Qiao, Balakrishna R. Iyer, Divyakant Agrawal, Amr El Abbadi, Sandeep Uttamchandani.
"PulStore: Automated Storage Management with QoS Guarantee in Large-scale Virtualized Storage Systems,"
IEEE International Conference on Autonomic Computing (ICAC) 2005.
Sandeep Uttamchandani, Guillermo Alvarez, Gul Agha.
"DecisionQoS: an adaptive, self-evolving QoS arbitration module for storage systems,"
5th IEEE International Workshop on Policies for Distributed Systems and Networks (POLICY 2004), pages 67-76, Yorktown, NY, June , 2004.
Sandeep Uttamchandani, Carolyn Talcott, David Pease.
"Eos: An Approach of Using Behavior Implications for Policy-based Self-management,"
14th IFIP/IEEE International Workshop on Distributed Systems: Operations & Management, 2003.
|
