Today EMC announced VFCache, aka Project Lightning, its entry into the burgeoning PCIe flash card market that so far has been the province of startups like Fusion-io, SolidFire, and Virident. While the startups focus on stand-alone systems designed for ultra-fast data capture, VFCache is designed to be integrated with EMC flash/disk storage systems at level 3 of Wikibon CFO David Floyer’s new model IT architecture model for the big-data/IO-centric era. Floyer will lead a public Peer Incite discussion of the implications of the announcement Thursday (February 9, 2012) at 12:00 p.m. EST (9:00 a.m. PST).
Flash memory is used so far mainly by online social media and service giants that need very high-speed, dependable data capture and access for big data. However, as more companies move into the big data capture and analysis area, this market is likely to expand to traditional enterprises.
Like the other PCIe products on the market, VFCache plugs directly into servers and provides storage above the IO software layer (layer 2 in the model), important because today that software is designed to work at slow disk read/write speeds. Like them, its largest attraction is very high speed operation. In a test in an Oracle environment it demonstrated a 3X improvement in speed and 50% reduction in latency, essentially operating as extra RAM storing data in persistent memory.
VFCache differs, however, in several important ways. First, it is primarily a read-only cache plugged into servers to provide fast access to data that is initially stored on EMC VMAX and EMC VNX, or on Symmetrix VMAXe, or EMC VNXe Flash-enabled storage arrays, some of which are hybrid integrated flash/disk systems (Layer 3 in the Floyer model). So while other PCIe products are focused more on capturing floods of big data, VFCache is designed for fast retrieval, for instance of video or still image files on a cloud service. However, VFCache can also be configured as a “split card”, which allows part of the card to provide fast, server-level write capabilities. EMC’s argument for this approach is that its customers are concerned about data protection and the possibility that data may be at risk on a PCIe card plugged into a server, even though flash provides persistent memory that will survive power outages.
EMC is also promising integration of VFCache with FAST (Fully Automated Storage Tiering) technology, within a year. This would provide an intelligent software layer that could move data between the server flash cards and EMC’s storage arrays automatically, based on access demand.
EMC also previewed “Project Thunder”, a purpose-built, low-latency, server-networked Flash-based appliance, essentially a set of PCIe cards in a separate, powered box, that will, it says, be scalable, shareable among multiple servers, and serviceable. It is promising an early customer access program for this product in the second half of 2012, will full release scheduled for 2013.
At Thursday’s Wikibon Peer Incite, leading industry expert and Wikibon CTO David Floyer will lead a discussion covering the implications of these announcements for EMC and its competitors, EMC customers and non-EMC customers looking for a storage architecture that supports PCIe Flash cards plugged directly into servers. The meeting is public and can either be accessed by dialing into the teleconference or by watching the discussion live on Siliconangle.tv.
For the past 20 years, functions like snapshotting, data replication and cache management have been moved off the server to the storage array. Now the pendulum is swinging back. The storage is moving back to the server to avoid latency penalties that come with spinning disks. This will require new sets of processes and procedures around such requirements as data protection and data recovery.
Latest posts by Bert Latamore (see all)
- How the cloud and IBM Watson are helping breed a better guide dog - May 3, 2016
- OpenStack Summit on #theCUBE: A window into the future - April 25, 2016
- Hadoop World: Big questions about Big Data’s future #HS16Dublin - April 22, 2016