Server/storage virtualization and a tiered storage architecture produced significant benefits for a large retailer.
By Dave Vellante and David Floyer
Note: This case study is based on a Wikibon.org community user, and while the case is real, the name of the company—CS1—is fictitious.
—While CS1, a large retailer, was satisfied with the performance and availability of its Tier-1 storage platforms (consisting exclusively of EMC's Symmetrix arrays), rapid growth of Tier-2 applications (mainly Windows-based) and increased demand for storage to support the company's growth were creating business constraints. Specifically, the time to migrate new arrays to support CS1's three-year leasing strategy was tedious and expensive (5+ months and up to $50,000 per array).
CS1 decided to adopt an aggressive virtualization strategy for both servers and storage to improve the flexibility and cost-effectiveness of its infrastructure. The company is nearly 12 months into a three-year plan to virtualize its entire SAN infrastructure.
CS1's original installation consisted of approximately 400TB of Symmetrix capacity split over two data centers, providing excellent availability and performance. Two SANs connected storage to more than 600 Unix and Windows servers.
CS1 leases equipment on a three-year basis. As a result, each year arrays and servers are moving out of, and into, the environment. This created three major issues:
- Time to migrate arrays averaged five months and required substantial planning because migrations necessitated shutting down applications for a period of time that was disruptive to business users;
- Migration costs exceeded $50,000 per array. Over the life of a typical high-end array this represented additional costs (above the expense of original hardware and software) of $0.43 for each $1 and;
- The current Tier-1 storage was too expensive to be a default for all data, implying a need for more granularity in storage tiers.
An assessment of storage residing on the Tier-1 arrays indicated that nearly 50% of the Tier-1 storage capacity was appropriate for Tier-2 placements, based on the company's recovery point objectives (RPOs) and recovery time objectives (RTOs).
CS1 was already a major proponent of VMware on the server side. The company embarked on a three-year project to virtualize nearly all storage attached to the SANs using IBM's SAN Volume Controller (SVC) virtualization platform to allow storage provisioning, migrations, and decommissioning without impacting applications.
CS1 has deployed IBM's DS4000 modular arrays behind the SVC to provide Tier-2 storage and has made it the default storage platform going forward. The company has retained the Symmetrix arrays for Tier-1 storage and is moving most of these behind the SVC. CS1 is migrating its array-based storage management software to Tivoli (see table, below).
Because the SVC is an in-band virtualization platform, CS1 had initial concerns that performance, availability, and data integrity would not meet user expectations. Careful testing had to be done on an application-by-application basis to ensure
- Performance of Tier-2 storage was appropriate for users conditioned to Tier-1 performance; and
- Placing Symmetrix arrays behind the SVC did not compromise performance, availability, and data integrity.
CS1 also created a new centralized group to assign data sets to storage tiers based on performance requirements. This group works with the lines of business to ensure there is an appropriate business justification for placing data on Tier-1 storage; otherwise, data is placed on Tier-2 platforms.
The project is 12 months in to a 36-month plan. Despite several hundred thousand dollars in hardware, software, services, and internal staff expense, the project is cash flow-positive. Payback period for the project was less than one year. For the storage it has virtualized (about one-third of the SAN-attached inventory), CS1 has reduced storage migration time from several months to five days on average. With virtualization of both servers and storage, CS1 has improved and continues to expect improvements in availability and time to provision a new virtual server with storage from weeks to minutes.
CS1 now estimates that with storage virtualization, for each dollar spent on hardware and software acquisition, the additional expense for migrating arrays has declined from $0.43 to less than $0.10. Moreover, the effective cost per terabyte for Tier-2-class storage has been reduced by a factor of three.
Despite initial concerns, performance, availability, and integrity of arrays behind the SVC, including Symmetrix, have not been an issue in production environments. However, the more-demanding Tier-1 applications have not yet been placed behind the SVC.
The figure shows the CS1 implementation strategy. The company wanted to continue to use the Symmetrix arrays as Tier-1 storage, but to move the majority of them behind the virtualization platform. The implementation plan is to virtualize as much of the storage as possible over three years. If the SVC cannot handle some of the high I/O requirements for Tier 1, CS1 has a fallback plan to implement a smaller pool of non-virtualized storage.
CS1's strategy is to make IBM's modular disk arrays its default storage devices. A special case has to be made by the business if Tier-1 storage is to be used. As there is no chargeback, CS1 had to ensure there were sufficient management controls in place to govern this policy. The company is very concerned that its users have sufficient performance as the storage is moved to Tier 2. To support this strategy, CS1 put in place a storage group responsible for placing data on the optimum tiers and for ensuring performance requirements are met by comprehensive testing.
CS1 is in the process of implementing a new performance and space monitoring and infrastructure monitoring mechanism, based on Tivoli software. The storage group monitors I/O performance and handles capacity planning and performance planning. The objective is to implement tools that monitor the performance, and if something is abnormal, to automatically open an incident report. The goal is to operate by exception and automation. For example, the Tivoli provisioning manager software is being implemented and scripted to automate commissioning and decommissioning of arrays.
CS1's long-term objective is to automate storage requests and storage migrations using a policy-based engine. The storage request would include the business requirements and prerequisites, and the policy would dictate where it goes, together with backup-and-archive criteria.
- Many organizations have concluded that front-end (server) and back-end (storage) virtualization are advantageous and often a prerequisite for an effective implementation of tiered storage;
- Virtualization is now ready for general adoption, with all major players putting significant muscle behind the technology;
- One of the largest benefits of virtualization is the ability to provision storage more quickly and migrate arrays in days rather than months. This reduces the costs of overlapping storage, delays purchase of new storage devices, and significantly reduces costs;
- A tiered storage strategy means that responsibility for the initial and ongoing placement of storage should be centralized; and
- Storage and server virtualization are independent projects, but users can gain significant benefits if they implement the two together.
CS1 has implemented a solid virtualization and tiered storage strategy, and the pressures for creating additional storage tiers will require an accelerated rate of adoption of virtualization and tiered storage. However a potential weak point of the strategy remains the degree to which CS1 can virtualize Tier-1 storage behind IBM's SVC, given the company's performance and availability requirements.
Dave Vellante and David Floyer are co-founders of The Wikibon Project, an open community of practitioners, consultants, and researchers dedicated to improving technology adoption. The authors can be contacted at email@example.com or firstname.lastname@example.org.
CS1's evaluation of IBM SVC proposal
- 2,000+ installations
- Quality references
- Ability to handle virtualization of installed Tier-1 arrays
- High-quality services
- Common software set
- IBM Redbooks best-practice guides
- Doubts about availability and performance of Tier-1 arrays behind SVC
- Concerns about SVC scalability
- Concerns about user perceptions of Tier-2 performance
- Friction from users pushed to Tier 2
Case study at-a-glance
CS1 is a retail company with an IT shop comprising mainframes and open systems. In all, the company has more than 1 petabyte of storage installed with about 40% residing on two storage networks supporting the company's mission-critical applications. SAN storage had been exclusively based on Tier-1 arrays, and CS1 needed a way to reduce the expense of storage acquisitions and lower complexities associated with migrations. One year into a three-year project, CS1 has saved several hundred thousand dollars and dramatically improved its time to migrate and provision storage. This payback came from a virtualized server and storage infrastructure that required careful planning and testing to ensure adequate service levels were met.