FusionIO goes public tomorrow. If you are an early employee, chances are, you are going to be a millionaire by the end of the day, at least on paper. The author of the great “fio” tool works there, and I hope this does work out for him and the rest of them well.
But … my question is a longer term one. Does the market … or will the market … support a higher cost PCIe channel flash as opposed to lower cost SSD based units? I’ve never satisfactorily answered this for myself. I can see a few divergent scenarios and markets where the PCIe flash are very important (and we work in these markets, and have a stake in them, so I want to see them succeed). But in the large scale general case … should we expect Flash in the PCIe channel or SSD in the disk channel to dominate local storage over time?
Ok, I am not saying spinning rust is going away. There are a few pundits whom are predicting this. The costs argue, strongly, against this. Storage density and capacity are going to be won by spinning rust for the near term future (2-4 years anyway). SSD 960GB units are available, and are about 50x the cost of the 1TB drives they would replace. So I don’t expect that to be occurring (the replacement) any time soon.
But … for some applications, the PCIe channel versions might not be as cost effective as SSDs, in that the latency advantages of the PCIe might not be worth the extra cost. Add to this the hot swappability of disks* and there are some significant advantages for the SSDs as compared to PCIe flash.
Basically, my concern is purely economic. Many vendors: Kingston, OCZ, Intel, … are in the SSD game, and will get economies of scale that the PCIe folks probably won’t get. Which puts cost advantages on their side. Further, the hot swappable nature, at least in theory, of some of these units would potentially render consideration of alternative non-hotswappable units moot. Which further limits the application space.
I am not saying its “Doom” for PCIe Flash. Just I am saying that the real markets for them are fairly niche focused (we are in some of the focused niches, so we have a stake in them). But we also have SSD product. 48x SSD’s in a 4U box. Yeah, pretty intense stuff. We’ve measured performance of these units individually and in RAIDs. Its pretty wild. 10’s of TB of SSD, for under $100k USD, and absolutely huge IOP numbers. What we do with these, apart from being awesomely fast IOP boxen at reasonable prices … yeah … thats what is getting interesting.
I was in the middle of writing this when I posted prematurely. Short of it is, yes, I think there is a market, but its narrow. I also think that Flash/SSD will have some meaningful competitors over time, and that there is a lower cost, higher volume side to the market. This lower cost side includes parts that are good, and parts that are junk. We’ve experienced both.
But there is a PCIe market for Flash/tier 1 type memory. This market is different, I think, than the market for SSD memory. They seem to overlap now, but i think this is simply poor segmentation. We need to refer to them differently to segment them properly. Which is why I refer to PCIe channel bits as Flash, and disk channel bits as SSD. Same underlying technology, very different mechanism to leverage.
I am guessing 2-3 companies will probably be able to handle 80+% of the market demand for PCIe flash. So FusionIO will go public tomorrow.
Moreover, as spinning rust companies contemplate their future, I’d argue that some of these likely successful vendors would probably be good pickings. Maybe not Fusion IO at this point, but Virident, Texas Memory Systems … Yeah, I think there are possibilities.
* ok, some SSDs are demonstrably non-hotswappable. You will, permanently, kill some of them if you pull power from them. Remember those useless Corsair units? We have a whole lotta them left courtesy of RMA replacements. Been experimenting, and sure enough, pulling power on them ala hot swap … permanently … destroys them.