Real comments on SSDs from the industry at large over at LinkedIn

It’s easy for pundits to flap their lips when speaking about SSDs. What’s harder is capturing what users, purchasers, and other influencers are thinking. That’s one place where social media really fits nicely into the picture. For professionally oriented discussions, I’ve found no place better than LinkedIn. Some of the group discussions are really quite thought-provoking. Case in point is this discussion about SSDs started just yesterday in the Semiconductor Professional’s Group and the comments are already quite informative:


At the moment the cost is the main obstacle to general adoption of SSDs. What do you think, how long that will last until we see reduction of price and increase of use for SSDs? -- Nima Firoozbakhsh

Bob Pownall, Ph.D • While I believe conventional hard drives will always beat SSDs on both the total capacity and bits of storage per unit cost metrics, I also think that SSDs will be "good enough" for many (perhaps not most, but many) users within the next two to four years. By "good enough", I mean a 256GB SSD with a cost premium no more than twice that of a conventional HDD. (Or at least at a price point below $150.) 256G SSD are already available, but they're more like 12x to 15x the price of a similar size HDD, with a price between around $500 and $750.

If a 256GB SSD was available below $150, even if it was 5x or more the cost of a similar sized HDD, I think many people would be attracted to the speed and perceived ruggedness.


Nima Firoozbakhsh • Bob, very good point. I also strongly believe that at the moment the gap is too big hence making it very expensive. Paying a premium for a better performance. Yes, paying like 12x more: not for mass production but rather a niche market.


Stephen Mitchell • I think there is a middle ground, a combination of a smaller SSD and HDD in a standard HDD form factor is a good compromise. You would partition both volumes such that the most used files are located in the SSD and the less used files in the HDD. Back in the good old days I used a RAMdrive driver in my XT to use some of my massive 640K memory. It contained my compilers and source files etc. in main memory configured as a volume and the rest in my 10MB HDD.


Nima Firoozbakhsh • Dear Alan, Stephen, to be honest I did not think about the ECC. But very valid and important point. Thanks for sharing...What do you all think about the following question: Will Flash SSD throughput more than double every year in the period from 2010 to 2012?


Justin Gedge • I like Stephen's idea. Several years ago [when building my current desktop for home] I wanted an SSD drive. With Linux- I can fit the OS and most software in 5GB. I was thinking about buying a 5/10GB SSD to load the OS and software. This would have been an incredibly fast machine... OS on an SSD and user volumes on traditional HD media [or network mount]... but at the time, you couldn't just walk into any store and buy them--they were still an OEM item. Maybe on the next machine I build up.

The issue of flipping bits- I hope they figure out how to get the error correction ironed out. As far as multiple writes, I can minimize this to when I install the OS and software in Linux and move the log files over to the traditional HD.


Alan Weiss • Justin: Error correction algorithms exist and are quite good, but it takes some good hardware architecture and good software inside of the flash device itself to do it properly. When I was contracting with Spansion as a Project Manager of a new high performance flash device, before they imploded (and have been resurrected, apparently), we were implementing 14-bit ECC. And you are correct, Justin: having software device drivers available in the open source world (and not just Microsoft) is important.

Nima: It is not just ECC -- it’s also the FTL that counts. That's software, and semiconductor companies are notoriously scrimpy (is that a word? ) on software expertise. By the way, hard disk technology is so advanced these days that I suspect it will have a very long life even if flash becomes more reliable. And no, I don't see throughput doubling every year between 2010 and 2012. I think flash memory busses need to undergo the kind of intense development that DRAM has gone through.

Stephen: You're right, more RAM == faster performance. At one time we jokingly said that there would be an intersection between 64-bit computing (i.e. a 64-bit operating system and address space) combined with vast amounts of RAM. We'd just mmap (memory map) the entire hard disk and be done with it! We were only kidding -- a little. It’s coming.
One of the main drivers behind using flash memory as your backing store is the potential to save energy. DDR3 and even more advanced DRAM technologies are coming at ever-decreasing geometries (e.g. 22 nm!) and density increases. Heat and energy become MUCH more important than even performance at a certain point. I believe that Google and others are at that point.


Just my 0x02 cents worth.