Redefining Unified Storage – Hardware Differences Part 1
by Hu Yoshida on Jul 16, 2012
Hitachi Unified Storage (HUS) has some unique hardware features, which differentiates it from other dual controller, unified storage systems. This is the second part of my series on Redefining Unified Storage, which looks into the hardware differences on the controller side. I will be following this with a post on hardware differences on the Unified Storage File Modules. Later I will have posts on the object/file differences, and the Management differences.
Dual Controller Differences
HUS, a dual controller storage system, is considered a midrange solution as compared with a multi-processor enterprise storage system like Hitachi Virtual Storage Platform (VSP). Despite this difference, the two share many of the same enterprise features.
These features include load balancing, dynamic provisioning, dynamic tiering, and synchronous/asynchronous replication. Competitive midrange systems can claim to have most of these similar features, but HUS is the only midrange system that can claim load balancing.
Load balancing is available in enterprise systems that have a global cache that can load balance I/O requests from multiple port processors and distributed them to multiple back-end LUNs. Midrange storage systems are limited to two controllers that have separate caches. This requires the manual assignment of storage ports and LUNs to an active primary controller while the other controller is assigned to be a passive backup controller. If this assignment of LUN ownership is not done and I/O requests are permitted to “load balance” across the two controller caches, the access of LUN images from one controller cache and that of the other would cause thrashing as the caches images are loaded and unloaded from the backend disks to the separate caches.
Dual controller systems without dynamic load balance require that performance constraining load imbalances on competitive systems must be manually diagnosed and then hot LUNS must be identified and remapped, followed by a manual re-mapping of the primary and failover paths across the SAN to the re-assigned LUN(s). True Active/Active controllers in HUS eliminate this disruptive process. While others may claim that their systems are Active/Active, they really mean that their controllers are both active, but on a separate set of LUNs. They do not have load balancing on the same set of LUNs.
Dynamic Virtual Controller
HUS is a cost effective , midrange storage system and utilizes two controllers with separate caches. However, unlike other midrange storage systems, these controllers can load balance the I/O activity and do not require the manual assignment of primary storage paths or LUN ownership. The rigid concept of LUN ownership by controller has been replaced with a more powerful method of LUN management. Now there is a dynamic global table of all configured LUNs that determines which controller will execute the backend portion of an I/O request for the LUN. This control list is independent of whichever front-end port is involved. This feature is called Dynamic Virtual Controller (DVC). While HUS does not have a global cache like VSP, it does have a global table. This provides enterprise load balancing across the two controllers for performance, QoS, and availability. This feature is described as a Dynamic Virtual Controller since it virtualizes the LUN assignment, in order to simplify the assignment of LUNs and dynamically load balance across the controllers and frontend port assignments.
DVC supports all native path load balance and path failover management such as Microsoft MPIO, Solaris MPxIO, Linux Device Mapper or IBM AIX MPIO for example. If a scheduled or unscheduled controller outage occurs, HUS will continue to run without loss of data.
Hardware Load Balance
DVC also includes a feature called Hardware Load Balance (HLB), which provides an automatic change to the controller management tables of one or more LUNs due to a sustained imbalance of the Intel CPU busy rates between the two controllers. This does not affect the mapping of the LUNs to the front-end ports (they remain unchanged). It only affects the assignment of the controller processes on the backend portion of the I/O request. While there are several complex triggers involved, a basic description would state that if one controller was exceeding a 70 percent average busy rate while the other remains below 30 percent, an HLB operation would take place. This may not result in a 50‐50 load ratio between the controllers, but it will be better than the larger imbalance that triggered the change. From an operations standpoint, DVC eliminates the tedious assignment of LUN ownership and the need to manually rebalance the LUN ownership when workloads change.
DVC is Hitachi Innovation, which differentiates HUS from other midrange and unified storage platforms. While other systems may use the same multi core processor and SAS technology, they do not exploit them in a way that makes them more efficient and cost effective.
Comments (6 )
That’s why I would love if “dual-active” term was more broadly used to describe storage with LUN ownership on dedicated controller. That would leave “active-active” term to truly a/a arrays like HUS.
I think the ‘dual-active’ shouldn’t be broadened because there is certainly some cross over between that and “active-active” – wouldn’t you say?
Thanks for your comment. I am struggling to try to differentiate what we do from what has traditionally been considered Active-Active or Dual Active. Remember when storage was first introduced with redundant controllers? The second controller was a passive controller that was only used when the “active” controller failed. Then vendors like IBM introduced the concept of active-active or dual active controllers where each controller could be active, but on a different set of LUNs to avoid the ping pong effect of loading cache images into one controller than another. This was a big selling point over the old active-passive dual controller storage systems. This is how dual controllers continued to operate for over 15 years until Hitachi introduced the AMS with the first dual controller storage system with load balancing across the active controllers. We are trying to redefine what is considered active-active or dual active. One differentiator is the fact that we can do automated load balancing. We have started to use the term Dynamic Virtual Controller, since we do virtualize the controllers from a LUN ownership perspective, but some people may think that implies external storage virtualization similar to what we do with the VSP. I welcome any thoughts that you may have to clarify what we do in the HUS.
Hu, this article does a very good job of succinctly describing the difference between active-active as used in HUS and the resulting affect on load balancing.
Does Cisco also use this dual controller in its system?
CISCO does not provide its own storage system, but if it uses the Hitachi Unified Storage, HUS, it can benefit from the automated load balancing that HUS provides. HUS can be connected to the CISCO server without the need to assign LUN ownership.