Enterprise or Modular, Enter the Enterprise Modular
by Hu Yoshida on Sep 10, 2007
For the past twenty years or so, there have been two basic types of external storage controllers, enterprise and modular. Basic Storage controllers provide the processing power to drive the port connections to the host servers, stage and destage data into cache to virtualize and minimize the physical latencies of rotating disk, provide the RAID mapping to physical disks, and drive the backend interfaces to physical disks.
Enterprise storage controllers consist of multiple processors, more than two, which can share the same cache image of disks in a common cache. This provides high availability since one or more storage processor can fail or be taken offline for maintenance, and the remaining processors can still access the data. More processors, means more port connections that an enterprise system can support and the higher the ability to support multiple failures. Performance can be enhanced by load balancing the I/O from the servers across multiple ports to the same LUNs. Advanced function Enterprise controllers like the USP V also have large, switch based caches which can support other functions like, distance replication, external attach of heterogeneous storage, and thin provisioning.
Modular storage is characterized by the fact that its controller can be contained in a rack which is usually 19” wide, and can support disks in modular drawers that are daisy chained to the controller in the same or adjacent rack. However, the most distinguishing feature is that, due to its modular packaging, its controller is limited to a cluster of two storage processors, each with its own cache. While you can buy modular storage with one processor, two processors are usually installed to provide write protection for data in the active processor’s cache until it is written to the disk. Since the caches are separate, a write to the cache in the active processor, needs to be replicated to the cache in the other (passive) processor to insure that data is not lost if the active processor dies. Running with one active processor creates an exposure for data loss if that processor also fails, so prudent storage administrators will stop and have the failed processor fixed before they resume production of critical storage applications. While other storage functions like point in time copies and replication are available on this controller, the lack of a common cache and limitation of processor cycles will impact production cycles, and so these functions are scheduled during off peak windows. For this reason, modular storage is not considered to be as available as enterprise storage. Each processor supports two or four storage ports so connectivity is also limited and load balancing is not done across the processors to avoid ping ponging the cache image between the caches.
Enter the enterprise modular storage, the USP VM which was announced today by Hitachi Data Systems. This is an enterprise storage controller that has all the features and functions of the base USP V, providing storage services to heterogeneous attached storage systems, supporting block, file, and content storage applications, but is packaged in a modular form factor,10 u high and 19 “ wide. However, instead of the modular limitation of two active/passive processors with separate caches, the USP VM has 32 active processors, that share a common cache. With 220 volts and 17.2 amps, it does not require special power or a raised data center floor. Essentially it has the packaging, pricing, and licensing characteristics of a modular storage system, without the limitations. It has all the enterprise features of the USP V such as virtualization of heterogeneous external storage, thin provisioning, distance replication, virtual ports, safe multi-tenancy and high performance front end 4 Gb/S ports and switched backend 4Gb/S disk. It can be purchased with internal disks or stand alone as a storage virtualization control unit. This is not to be confused with virtualization SAN appliances which do volume management in the SAN. The USP VM is a full function storage virtualization control unit that does more than volume management. It also is not dependent on the SAN and can support mainframes as well as NAS and Content Archive applications. With all these capabilities, the USP VM is far superior to traditional modular storage systems like 3ParData, Datacore, FalconStor, EMC CX, and LSI, as well as virtualization appliances like SVC and Invista. The USP VM is not meant to replace all modular storage but to enhance and increase utilization of existing modular storage with enterprise storage services. One USP VM sitting in front of a number of heterogeneous modular storage systems, can provide enterprise class services like application load balancing, non disruptive maintenance, distance replication, non disruptive migration, policy based tiered storage, storage virtualization, thin provisioning, and more, using a common set of management tools.
With 1.2 Million IOPs of front end performance it also competes very favorably with the latest DMX 4 with the added advantage of storage virtualization and thin provisioning with less power and cooling requirements. It can also be configured with the USP V for enterprise core and edge requirements, the USP V in the core data center and the USP VM in the distributed branch office or remote vaulting site. Its virtualization and replication capability also makes it an ideal migration tool for data migration with in a data center or replication to new data centers. The compact packaging and low power requirements may also enable it to be a mobile migration tool for consolidating or redistributing data.
Comments (4 )
Your blog points out that for the last 20 years there have been two types of storage controllers, enterprise and modular, where the former consists of multiple processors and the later of just two processors. 3PAR, you say, is an example of a “traditional modular storage” system.
As a point of clarification, for the last 6 years, 3PAR has been shipping a third type of controller architecture: massively scalable clustered modular. 3PAR InServ Storage Servers consist of 2 to 8 cache-coherent and autonomically load-balanced processors. These processors are added non-disruptively and, once added, take on their “share” of each and every workload automatically. This architecture allows service levels to remain high and data to be fully protected, even under processor failure conditions.
Thank you for this opportunity to clarify matters and avoid any confusion for your readership.
Director, Product Marketing
Your blog states 1.2 million IOPs but both the HDS and Sun 9985V datasheets say 0.8 million. It is not mentioned in the HP XP20000 datasheet.
When I use application block size as “1 MB”; I found that two controller in AMS 200 giving dissimilar values (here I created two RAID groups and two 10 GB LUN (each LUN configured for 256 KB LUN stripe size) and assigned to the individual controllers with cache partition size as 100 MB with 16 KB segment size; is there any upper limit for application block size for getting maximum performance in AMS 200.
[...] a baby brother known as the USP-VM. This was a cut down deployment of the USP in a class that HDS referred to as “enterprise modular”. Prior to that, Hitachi offered the NSC55, which at the [...]