HET would better called MET
Or SHIT for Seriously Horrible Intel Technology.
Intel's delayed 710 SSD, its X25-E replacement, has arrived at last, after having been initially outed back in July. It's a fairly standard – for these days – 25nm, 2-bit MLC solid state drive, with 100, 200 and 300GB capacity points. But it has wildly skewed read and write performance and poor-to-average bandwidth numbers. …
Many years ago, I worked with a delightful gent who told me his first job as a student was "vavle monkey". That entailed working with one of the original computers back in the '50s, his job being to sit all night in the boiling computer room with a box of valves, ready to leap up and replace any blown valves. He said he could go through a dozen valves in a night, as long as he didn't accidentally electrocute himself in the valve changing process (that could shock you unconcious!). Obviously, Intel is doing the economy a favour and turning the clock back to those good 'ol days, providing employment for thousands as "SSD monkeys" (not to be confused with "tape monkeys", who switch backup tapes for a living).
Tne now-defunct Wang Laboratories came up with a model 710 disk, too -- 256K fixed and 256K removable, ca 1973. The hardware and electronics were a phenomenal piece of engineering mediocrity, even for that time. One could have done about as well trying to make a disk drove out of a phonograph turntable with the tone arm recycled as head positioner. But,unlike any disk today, it was byte-addressable.
The Intel 710 article brought it all back in a rush.
I wonder if they retained the Wang feature of electrostatic formatting. Walk across a carpeted floor, touch the control panel bezel, and watch the drive auto-format both platters.
When the original x.25 came out in 2008, I held the SSD World speed record for a couple of months at 39,600 Random Read IOPS. The value of SSD is that huge RR IOPS number, together with a good (50%) random write IOPS, at least out of the box. $/IOPS is important provided the IOPS number is in the ball[park of the leaders.
SSD is a rapidly evolving business, with plenty of players. Failing to follow something like Moore's Law is a bad thing.
I must add, however, that the PCIe accelerator class products can beat discrete drives to their knees, performance wise. If local SSD is considered more as a DRAM extender than a disk drive, this is where the action will be. With storage appliances architectures moving to x86 COTS motherboards, the PCIe approach looks to me like a better place for Intel to focus. Perhaps a direct QPI connect flash device might be an even better fit for a storage optimized design.
Biting the hand that feeds IT © 1998–2020