Flash Revolutionizes IT
November 27, 2012, Server Design Summit, Santa Clara, CA—Gary Smerdon from LSI described the data deluge and other drivers in creating an inflection point in the data center storage hierarchies. The increases in technologies and in data volume are driving advances in the data center that are also changing networks and communications.
By the yeare 2020, the world will have over 50B devices generating and communicating data. The data growth of 30 to 50 percent per year is now starting to stress storage and network capacities. The biggest growth has been in the cloud and mega-data centers due to Web 2.0 types of interactive content and cloud services. In addition, the high performance computing segment is adding velocity and reliability to the list of requirements for the data.
As a result, the needs of the mega-centers are driving innovation and change in systems architectures. All of the big systems are on a massive scale with homogeneous server types for modularity and increased automation. The open-source software and on-demand cloud capacity have enabled a network that is optimized for operating and capital expenses.
Among the challenges for the big systems are analysis and extraction of value, data protection, system performance, and the depths of the user experience. Operations are getting more complex and costly, generating concerns about total cost of ownership, life-cycle management, resilience and manageability, and capital costs and power.
All these changes have exposed a gap in performance between the serves and the storage. HDD latency is now the biggest part of the data use budget, and the drives cannot get any faster because they are RPM limited. HHDs are the storage medium of choice for large capacity and lowest cost per GB. Flash can fill the gap in cost and performance between the HDDS and the rest of the system.
By the year 2015, flash is projected to become a $500 M market for primary storage on PCIe for the early adopters. It will be $1.48B for cache functions, and direct attached storage acceleration will be over $3 B. There will be optimized solutions for various segments that will balance performance and costs.
As flash is adopted in the enterprise, the increased performance is quickly noted, but the high number of writes in a typical workload is calling for more details on durability and retention. The solution for this issue is to have closed-loop measurements of each cell and enough processing power at the storage to address wear leveling and retention thresholds.
Advanced error correction coding and signal processing as done in the HDDs will provide the needed durability and endurance. Flash offers the opportunity to bridge the gap between memory and storage with increased agility, high availability, and good performance under failure. The continuous address of a flash memory, when compared to the fixed sectors of a HDD, allow for graceful degradation with failures. When a HDD loses a sector, the OS receives a failed seek and tends to crash. When a SSD starts to go bad, the controller can just reassign the address to a spare block and OS doesn’t see any chagen expect for a changed byte count. Flash is now part of all the new architectures.
