Virtualization Redefines Tier 1 Storage
by Hu Yoshida on Mar 14, 2012
When you Google the definition of tier 1 storage you get a variety of results.
In older definitions of storage tiers, it was all about the media—tier 1 was expensive hard disks, while tier 2 was optical disk and tier 3 was tape. No consideration was given to the control unit functions. Today that has changed…
The best definition I found for tier 1 storage today comes from David Floyer of Wikibon.
Wikibon defines tier 1 storage as:
- Has the time-tested ability, through whatever array resources are required, to make the critical application perform at the highest possible levels required for the business
- Has a complete and fire-tested set of high availability, remote replication functionality allowing large-scale consistent replication to multiple locations
- Has a set of well-established performance and availability services that understands both the technology and how to integrate and project manage that technology to meet the specific organization and industry imperatives, and prove those requirements have been met
- Can support mainframes as well as open systems (which continue to run a large percentage of mission critical applications, especially in the financial sector)
In order for critical applications to perform at the highest possible level at all times, the storage system must have a multi-processor controller with a global cache, which enables it to load balance across the processors and continue running by switching around processor and cache module failures. A dual controller storage system, with separate caches, cannot be considered tier 1 because the loss of one controller and cache will cause a degradation in performance and the risk of data loss (if the second controller should also fail). A cluster of dual controller storage systems, connected by an external switch, cannot solve this problem since the application is connected through one dual controller cache node in the cluster.
High availability, Remote Replication
High availability, remote replication to multiple locations—meaning three or more for out-of-region data recovery without data loss—requires that the control unit have the processing power and bandwidth to support this extra load. Again, a dual controller system that must do all these functions (in the same controller and cache) cannot support this requirement. The other controller and cache are only there as a passive standby.
Performance and Availability Services
The third point in this definition addresses the need for an intelligent controller that can optimize the functions like tiering, thin provisioning, snapshots, replication, and migration to business requirements, and provide a show back to the application or business user. This involves support of application APIs, plug-ins, adapters, and client/providers (like SMI-S) and a tuning or scheduling capability based on utilization and input from the application.
The support of mainframes and open systems requires control units that can support CKD, as well as fixed block records from open systems. In order to support both at the same time, the storage controllers must also be able to partition resources so that I/O intensive mainframe channels do not dominate the cache and processors at the expense of the open systems I/O.
As we look at this definition, we quickly realize that today’s tier 1 storage is about the functions and capability of the storage control unit, and not about the media.
Now let’s take this one step further and separate the tier 1 controller from the media so that it can attach other storage systems, like modular dual controller storage systems. When we do this we essentially have a tier 1 storage system, even though the systems that we attach, or virtualize, behind the tier 1 controller are not tier 1 storage systems.
So what is a tier 1 storage system today? It is any storage system that sits behind a tier 1 storage controller, or is virtualized behind a tier 1 storage controller. In other words, you don’t have to buy tier 1 storage to get tier 1 storage capability. As long as you can attach it behind a tier 1 storage controller like VSP, you can upgrade your existing storage resource to tier 1 storage.
Comments (4 )
What about level of support from the vendor? Does that not factor into a Tier 1 definition?
Hu, good blog. But does this make the HUS VM Tier 2 or Tier 1.5?
(Disclaimer–I work for a competing storage vendor.)
HUS VM has all the functions of the VSP except for mainframe support, so it does not fit David Floyer’s definition. However, I would not classify HUS VM as what some people describe as tier 1.5. Tier 1.5 is a colloquial term for storage that fits between a tier 1 global cache storage system and a tier 2 dual-controller storage system. Wikibon describes tier 1.5 systems as controller designs that are loosely coupled for scale-out requirements. http://wikibon.org/wiki/v/What_are_Tier_1.5_Arrays
HUS VM is not a loose coupling of processors with separate caches, it has 16 processor cores that share a global cache. Therefore, I would consider it a tier 1 storage system that can virtualize tier 1.5 and tier 2 arrays and enable them with tier 1 scale-up and scale-out capabilities. It is an enterprise tier 1 storage array that is packaged and priced for the mid-tier market.
Azim, you are absolutely correct. Vendor support is essential for tier 1 enterprise storage systems. The readers of Storage Magazine ranked Hitachi Data Systems’ Enterprise storage system, number 1 in technical support as well as number 1 in Sales-force Competence, Initial Product Quality, Product Features, Product Reliability.