Hybrid Hard Drives Just Need 8GB of NAND

judgecorp writes "Research from Seagate suggests that hybrid hard drives in general use are virtually as good as solid state drives if they have just 8GB of solid state memory. The research found that normal office computers, not running data-centric applications, access just 9.58GB of unique data per day. 8GB is enough to store most of that, and results in a drive which is far cheaper than an all-Flash device. Seagate is confident enough to ease off on efforts to get data off hard drives quickly, and rely on cacheing instead. It will cease production of 7200 RPM laptop drives at the end of 2013, and just make models running at 5400 RPM."

Comments

Re:Moving parts and fatigue

[tlhIngan]

The lifespan issue with SSDs has three main factors.

1: Type of flash memory (SLC, MLC, TLC, in order of decreasing durability)
2: Size of the flash drive (larger drives have more room for wear leveling algorithms to work with, thus staving off flash cell burnouts due to exceeding maximum number of writes).
3: The amount of throughput on the flash drive. An expected heavy load is roughly 10GB/day. Doubling the load halves the lifetime of the drive. Quadrupling the load quarters it.

Granted, the cache on a Hybrid is being used a bit differently than how you would use a straight SSD. But, with such a small cache drive, you ARE going to wind up cooking it after a relatively brief period of time.

Which for most users and usage scenarios, is basically forever. There's been a volunteer-run test of longevity which stresses an SSD until it fails by writing data to it continually. And the SMART data typically gives you plenty of advance warning - the Media Wear Indicator (MWI) tells you how many cycles are left in the array - once it hits zero, it means the number of write-erase cycles has hit the guaranteed limit and you're running in unknown territory (though there are usually still spare blocks and most will still have plenty of life). If you want guarantees, once the MWI hits zero, it's time to back up and get a new SSD. The tests run until the drive itself dies which tell you how long you have left. So you generally have a LONG indication of media wear out.

However, the biggest problem SSDs face is actually sudden loss and corruption of the FTL tables (the ones that map logical sectors to actual flash sectors). If you hear of SSDs dying prematurely, it's almost always because of table corruption. These tables contain things like sector translation, sector wear, dirty/clean bits, trim status, etc.

In the past, you could regenerate the tables from the spare area data (typically 16 bytes per 512 byte data area), but use of enhanced ECC algorithms consume that space up to accommodate better error handling. Plus it also meant way longer mount times as the controller had to scan the entire media for the information (many seconds long).

These days, controllers come with 512MB or more of RAM to hold the tables in memory for quick access. The problem is the tables are often written out lazily to storage, which means if you yank the power suddenly, the SSD might not be able to write the dirty data to stable media, or worse yet, it'll be in the middle of the write operation which leaves data in an unknown state.

Good SSDs often have piles of capacitors to serve as emergency power which can keep the array powered for a couple of seconds - more than enough time to flush the tables to storage and protect your data. Of course, this costs a lot more money and is usually present only in the top tier drives and enterprise class SSDs. If an SSD dies suddenly, it's usually because of this.

Hard drives use the back EMF produced by the spinning platters to perform emergency shutdown procedures, including retracting the heads.