Ed. Note: This week, we introduce our newest column, Hard-Core Hardware, which replaces Server Snapshots.
The reason is obvious performance. With HDDs achieving maybe a couple of hundred operations a second (ops/s) at best, RAM providing somewhere in the vicinity of tens of millions of ops/s, and CPUs providing billions of ops/s, the disk has been a major bottleneck for decades. Companies like Diskeepe (Burbank, Calif.), have even made healthy livings eking more performance out of stuttering hard drives by addressing rampant fragmentation.
Now factor in dual-core, quad-core and higher-core processors as well as better memory technology in general and the gulf between the HDD and the rest of the machine just keeps getting wider. SSDs are a step in the right direction.
According to Dave Donatelli, head of information storage at EMC (Hopkinton, Mass.), flash beats the highest-end disk drives in Symmetrix DMX arrays by a factor of 30 in terms of Ops/s. That's why EMC has adopted SSDs by STEC (Santa Ana, Calif.) for its DMX systems and will be adding them to its Clariion line.
Fibre Channel (FC) disks are faster than the drives found in everyday servers. Yet the best FC disks can do no better than 6 ms in response and that's if traffic is light. When the workloads spike, performance falls off. Flash, on the other hand, provides access rates of 1 ms or less, no matter how much traffic there is. Some types of SSD can actually provide access rates of around 0.01 ms that's 250 times better than regular hard drives.
While plenty of storage vendors have already jumped onto the SSD banner, HP (Palo Alto, Calif.), is planning to use flash in its servers in the near future. Initially, of course, this will be probably be in heavy-duty systems such as the Superdome and Integrity, as the technology will come with a definite price premium. But that will change rapidly.
In fact, a recent report from IDC (Framingham, Mass.), puts the SSD market at a healthy $400 million per annum, and it's expected to grow more than 70 percent per year for the next five years. That adds up to $5.7 billion by 2012. In comparison, the entire HDD market currently does around that in a quarter.
So I'll go out on a limb and say IDC is grossly underestimating the growth of SSD. Why? Ever heard of a subject called green IT? Much as I hate to mention it, this is the buzzword du jour. "Green" has captured the imagination of those at top management levels, and it's not going to go away for quite some time. Those guys up there have finally cottoned on that IT wastes a lot of power. They are right to demand a solution. SSDs hand it to them on a plate they use about 98 percent less power than regular disks. So it's only a matter of time before the green-eyed monster virtually outlaws the use of power-sucking hard drives.
Certainly, the price will have to come down substantially. And it will. Watch out for startups galore in this area attempting to pull the rug out from the established players with new designs or components that bring the price way down, boost performance even higher and tackle unresolved issues, such as density. Currently SSD can't compete with ATA for sheer volume of storage. In addition, things like buses will have to become a whole lot faster so they don't create another bottleneck.
One final reason for the hard drive's decline will be the amount moving parts. Hard drives are the weak link when it comes to failure rates in existing servers and storage arrays. Remove hard drive failures from the equation and the lives of system administrators become much easier. SSDs have no moving parts. Their failure rates are far lower than spinning disks.
Within a year or less and a small server vendor will likely come out with SSD-based servers unless HP gets there first. The rest will follow suit. It will be a price-premium option at first, then it will become standard; the only question is how long it will take. I give it five years. And within two, some of you will be using PCs and laptops with flash drives inside.