August 13, 2013, Flash Memory Summit, Santa Clara, CA—Zahid Hussain from EMC talked about the datacenter moving to more solid-state storage. The evolution to hyperscale datacenters is driving changes in architecture and equipment.
Adding flash into the storage mix is not the same as adding another tier of memory and storage. The hyper-scale datacenter requires performance, power efficiency, agility, and simplicity. The changes in the applications has also changed the expectations for all components in the system.
The platforms have changed from mainframes in the '80's to client-server architectures and now to mobile and cloud. All phases of the hardware have increased performance by 10-1000 times. The issue is that CPU and memory have increased much more than the HDDs. The great benefit of flash is that it fills the growing CPU-HDD gap.
The system architect still needs to have good models of the hardware and software to define the needed changes. The applications define the storage requirements and performance, leading to all flash arrays from today's hybrid storage structures.
For the low-latency functions, PCIe flash and specialized management software to pool and aggregate storage nodes is needed. This set of tools can tier across all storage and other requirements to support various categories and applications.
On the server side, the market for PCIe and SSD storage is in the $1.5B range, and is expected to grow to $5B in the next four years. The software models are helping drive adoption to structure high-density, high-performance, and acceptable power in storage arrays. The software at the server enables another layer of cache in a direct-attached store with scalability, coherency, and resiliency.
Systems architects also have to scale I/O to meet the system needs. Pools of server flash can aggregate the total datacenter output requirements in dynamic storage tiers. For extreme levels of I/O, an all flash array facilitates software-defined storage and is inherently balanced. The tools enable flash-optimized storage for content-based architectures and can include in-line de-duplication as an integrated function. The tools can translate through the mapping layers for all operations for integrated performance and other functions.
The latest tools are linearly scalable for performance and capacity with all services. The new capabilities require changes in architecture and changes in features and functions for optimized performance. The shift to solid-state storage increases the rate of change in the systems, enhanced by the range of flash products. Selecting the appropriate flash product depends upon the application.
