Storage Validation
September 18, 2013, Storage Developer Conference, Santa Clara, CA—Amritam Putatunda from Ixia wants us to prepare for the coming data storm. The data infrastructure needs to evolve to keep up with the growth in data types, complexity, and volume.
Network infrastructure validation is a maturing model that will help manage network infrastructure. Increasing complexity makes validation and optimization challenging. Pre-deployment validations include functional and performance modeling and testing, deployment validation tests the integration and optimization of the various system components, and post-deployment helps with change management and security.
In most data centers, the network infrastructure gets much more tension than storage infrastructure. Network infrastructure has strong validation frameworks, and substantial efforts identify blocking points in the networks. Until now, storage infrastructure has been restricted to basic go no go testing, and still everything is expected to work.
No one doubts the need to test network infrastructure. Applications grow in complexity and changing threat landscapes require networked validations. The infrastructure is checked for application delivery controllers, load balancers, deep packet inspection, firewalls, data leakage prevention, and application aware routers and switches.
These network checks only big the question, is only the network infrastructure that needs to evolve? Even though the storage infrastructure has been evolving to match the changes in Ethernet, storage is still considered just another add-on component. The reality is that hardware innovation in storage has been growing.
Now storage can be configured as scale out and scale up NAS technologies, which helped reduce cost through continuing innovations. Breakthroughs in storage technologies like flash solid state drives are changing the price-performance metrics, as are changes in array architectures.
IT managers can no longer just throw more hardware at the problem because the answers are too complex for a simple hardware fix. Without a proper and strong validation framework, evaluation of variables for costs, reliability, change management, etc. are very difficult. Validation requires application data pattern emulation which will differ based upon the applications, access modes, and data types. Standard benchmarks only measure nominal workloads and other parameters, which may have no relation to production data requirements.
Data pattern emulation tools can eliminate additional variables from the eight environment, calculate maximum performance of storage per application, identify and optimize delays, and isolate issues in configuration and management. These tools can also perform selection procedures to screen vendors for best fit to a particular environment.
The steps in application emulation start with classification of the data types and workloads. Once the classifications are generated, behavioral analysis target application specific block and file size distributions. The analysis also analyzes sequencer one random patterns in the I/O and inspects application-specific data patterns to provide realism in the tests.
The next step into the process is protocol and user behavioral analysis. Their are many parameters here that are not necessarily independent. Some of the analysis targets are control plane, caching abilities, locks and leases, cueing efficiencies parallel file system accesses, access controls and access behaviors, all within a dynamic user behavioral environment which affects load patterns as well as issues of concurrency, reliability, and resilience.
An example of a application-specific test is video over SCSI. A video right generally happens over a fixed block size irrespective of data content. This creates holes throughout the disc, so a video server must be aware of the gaps for maximum scalability. In comparison, predictable, analyzable data generation needs analysis efficiency measurements that map to analyzable realistic data patterns. These patterns are often stored and distributed file systems, which provide us with an efficient storage infrastructure for both structured and structured data.
Pattern emulation for cloud applications can be a difficult challenge. Application behavior is largely unknown since the cloud infrastructure may mask storage issues. It is quite common for applications on the cloud to fail due to mismatches between requirements and capabilities. Many running apps have never been tested with real traffic and users.
Networking and storage infrastructure technologies need to evolve together for optimal system response. Validation frameworks can ease the integration issues associated with constantly changing network and storage infrastructures. Testing and validation must use workloads and data sets that are representative of the production use or they are of little or no value.
