Solid state drives – flash memory that mimics very fast hard drives – are coming into their own in data centers as prices drop, reliability increases and more CIOs become aware of how the drives can enormously speed data access.
SSDs remain niche products, primarily because of price (a Fusion-io 640-gigabyte drive costs around $10,000). But because of higher memory bandwidth (100 times higher or more than a typical drive), they can be extremely cost effective when used for databases, data warehouses or scientific applications, or if super-fast access to data is critical and an app is storage I/O bound.
For instance, working with Samsung's NAND flash technology, Fusion-io has created ioSAN, which is capable of replacing an entire enterprise storage area network using a single PCI Express card with integrated network ports.
Other manufacturers of flash SSDs include Hitachi Data Systems, Intel, SanDisk, STEC, Sun Microsystems and Toshiba.
Number of I/O operations per second that Fusion-io and Hewlett-Packard achieved this spring running five 320GB ioDrive Duos and six 160GB ioDrives on a quad-core ProLiant DL785 G5 server, sustaining throughput of 8GB per second.
David Flynn, Fusion-io's chief technology officer, points out that most databases are less than a terabyte in size and so the need for I/O performance often leads to over-provisioning servers, disks and RAM to get enough storage bandwidth.
“One of the key indicators,” Flynn says, “is you've got lots of servers in an array, and you find that any one of them is only using a small amount of CPU. You've got an excess of servers, an excess of RAM and an excess of mechanical spindles.” In a case like this, switching to properly applied SSDs can save on the number of servers, the amount of RAM, the number of spindles – and the cost of software licensing. Why? The answer is simple: Fewer servers mean fewer licenses.
There are also operational savings. SSDs require less cooling and use less energy, so the operational cost is low compared with an array of hard disks, especially the 15,000-rpm drives used to get maximum performance. The result is that in spite of the high costs of SSDs, a properly designed SSD installation can save money on total data center costs.
The interest in these drives is growing in enterprise environments but is still relatively untested, says Joseph Unsworth, a research director at Gartner. A chief reason, he notes, is a lack of standards.
Unsworth warns that not all SSD products are created equal. There are many manufacturers, no well-established standards and enormous variation in reliability, speed and other critical factors.
“There are hundreds of SSD vendors out there all claiming to have tremendous solutions,” he says. “But there are only a handful of companies that can truly deliver.”
A Question of Life Span
Each flash memory cell can be written to only a limited number of times. This is called “write endurance,” and it used to be the limiting factor on the life span of solid-state drives.
But write endurance with modern flash SSDs is a nonissue. It takes decades of constant use for writes to affect a flash SSD. A combination of improved technology with much greater write endurance makes this much less of a problem than it was in the early days of flash.
SSD makers employ load leveling, in which the number of writes for all memory blocks is equalized, and extra blocks of memory can be swapped in for failing blocks.
The more advanced makers employ additional techniques. For example, Fusion-io organizes its flash memory in a RAID-like fashion for built-in redundancy.