Related topics

OCZ's automated flash-HDD tiering for PCs

Like EMC's FAST but for consumers

OCZ is extending its RevoDrive PCIe flash card to deliver automated hot data placement on solid state storage from a direct-attached SATA hard disk drive, using adaptive cache storage management software.

The idea, presented at the Flash Memory Summit, is to extend the memory hierarchy in desktops by integrating a PCIe flash memory card (SSD) with a hard disk drive (HDD) and using NVELO's DataPlex software to detect and automatically move hot data from the SATA HDD to the flash cache.

NVELO says flash is too expensive to replace a hard drive but is darn good as a cache for disk. This, in miniature, is what NetApp is doing with its Flash Cache, previously call PAM (Performance Acceleration Module), and what Seagate is doing with its Momentus XT hybrid hard disk drive which has 4GB of on-board flash.

The DataPlex software executes on the host PC and its algorithms, using the SSD as as a HDD cache, store the most frequently and recently used data on the SSD. There is no user action required to pin files in the SDD. NVELO says this is the way to get affordable SSD performance without blowing your budget on a compete HDD replacement.

It reckons the cache should be four times the PC's memory capacity and that, because its software runs on the host PC, it gets block and file-level visibility into system traffic with insignificant host CPU processing needs.

NVELO has a chart showing showing how its software using an SSD cache produces a SYSmark productivity suite score that's less than ten per cent below a full SSD system, being much faster than a Momentus XT and a Hitachi 7200rpm, 2.5-inch HDD.

Price-wise NVELO talks generally of a 320GB HDD costing $50, a 160GB SSD costing $350 and its 320GB SSD + HDD + software combo costing $150 providing "half the price, twice the capacity [and the] same performance" as an SSD-only alternative. Another way of slicing this cost/GB pie is to say DRAM is $20/GB, NAND flash is $2.00/GB while HDD is $0.20/GB.

RevoDrive-B

OCZ is introducing a RevoDrive-B which uses the DataPlex software to load its multi-level cell (MLC) flash with data. A SATA HDD can be directly connected to the RevoDrive-B PCIe card and Windows and Linux hosts just see a HDD which happens to have a transparent SSD cache in front of it. OCZ calls this a consumer tiering product and says it can be interposed between an PC and its existing SATA HDD as a generalised disk I/O accelerator. OCZ explicitly likens this to EMC's FAST (Fully-Automated Storage Tiering), saying it's the consumer equivalent of that enterprise set-up.

This is, in terms of memory hierarchy evolution, a near-term architecture with the far term possibly being to have a PC with its processors and their cache sitting at the top of a hierarchy. Below them is not DRAM system memory but a DRAM system cache sitting atop flash system memory, and this is layered above a flash cache sitting in front of HDD storage.

We've used the term "flash" but only for convenience. It's really any non-volatile memory, meaning flash or a post-flash technology with improved performance, durability and energy usage. For this kind of memory hierarchy to emerge then operating systems will need to take the architecture on board. With Microsoft's glacial pace of Windows development that's unlikely to happen soon. Apple may, "may", add support for this kind of memory tiering in its next major Mac OS X release. Wouldn't that give it a neat performance boost compared to Windows?

It's possible that server manufacturers also have this kind of memory hierarchy scheme in mind for future server development, particularly the ones using PCIe flash accelerator cards, like Dell, HP and IBM.

PC users for now will have to be content with retrofit products like the RevoDrive-B, whenever that comes out. Since OCZ is now talking about it we'll estimate it's a first half 2011 product.

Thanks to PC Watch for selected slides from the Flash Memory Summit. &Reg;

Sponsored: 10 ways wire data helps conquer IT complexity