Hu's Blog

During the past week I came across some systems that claimed to be hybrid storage controllers. When I looked into what they were doing I saw that they were primarily addressing the use of SSD Flash Drives in combination with HDD or spinning disk drives.  While they are combining two different types of technologies and maximizing their use with automated tiering so that the net effect is an improvement over using them separately, I believe that this is more an example of hybrid storage capacity rather than hybrid storage controller.

A Hybrid Storage Controller needs to be the combination of hardware and software controller technologies. Hitachi has taken a Hybrid Storage Controller approach within our enterprise Hitachi Virtual Storage Platform (VSP), our high performance Hitachi NAS, and our Hitachi Unified Storage (HUS). Hybrid in Hitachi storage controllers means the combination of multicore processors which best address complex management tasks with hardware ASICs or FPGAs, which are used for acceleration of front and backend data movers.

Hitachi VSP was designed with a separate pool of Intel multi-core processors to handle complex management tasks like dynamic tiering and replication, while data movement through the front and back-end ports is done by customized ASICs. This hybrid architecture relieves the memory and bus bottlenecks that result when the processor software has to handle all these functions including the movement of data. This architecture has a modular design, which enables the controller to scale up, out, and deep through external virtualization by adding an incremental processor, a port ASIC, and cache modules. This separate pool of multi-core processors becomes more important as applications like VMware continue to push more functions like VAAI down to the storage and demand more information from storage through VASA.

The following figure shows the basic hybrid architecture of VSP. The VSDs are multi-core Intel processors, which run the management software for functions like dynamic tiering. The FED and BED are ASICs, which handle the data movement, connection, and backend RAID.  They are connected through an internal switch to a global pool of cache modules. This figure shows only one switch plane. Additional layers can be added to this switch configuration to non-disruptively scale this to a total of four layers of processors, ASICs, and cache modules.

Hitachi NAS and the HUS file modules use a combination of multi-core processors and hardware FPGAs to address the high performance and scalability required to address the demands of big data.  In my next post I will describe the hybrid architecture of the file modules in HUS.