Is the all-flash data centre just a tantalising dream? An all-flash data centre? It’s an intriguing idea. The hardware and software components may be there, but what about the business case, including the overall price/performance and total cost of ownership? Yes, such a data centre will be free of power-gobbling and rack space-consuming spinning disk enclosures, bringing …
"taking disk drive arrays out doesn't save a square millimetre of space; it is still there, just empty, and the air in the data centre still needs conditioning"
It may need conditioning, but that few square inches (feet or metres depending on your footprint) will not need *cooling*. And of course, taking the array away means a reduction in power which, if your racks are metered, can lead to a tangible saving.
The article headline is nonsense, but the body of it does talk sense - I expect DCs to be a mixture of SSD and spinning rust until such time as SSD costs come down. We'll see a gradual progression to SSD but also, possibly, less big SAN footprints, as the direct access to large storage volumes does away with the fibre requirement.
Depending on your environment, it is quite possible that in the region of half of your storage requirements are for backups. If you are backing up to disk, then the cost for flash is likely to be pretty high when compared to a disk based solution. Space savings are minimal if you are looking at high density disk arrays, and in many cases the flash equivalent will actually need more physical space.
Been running all the front and back end servers for my organization on Intel SSD's for a couple of years now. Couldn't be happier with the power consumption, speed and reliability improvements.
I would suggest any organization looking at doing it this way, has a good handle on what data they are actually storing and for how long. Its amazing the efficiencies that can be done when you are actually restricted to storage space.
If you consider the number of man years spent in design/build/test, I think you'll discover that the majority of the effort is in the software vs the hardware, even if the HW is fully custom. Next, you'll want to do the analysis to find out exactly how much you're winning by doing more custom HW vs off the shelf HW. It's considerable, but not nearly as much as some of the HW centric folks would hope for. So, a fully mature company would probably have both, but you can harvest the important custom HW portions later on in a piecewise fashion.
After all, things change fast enough & volume is low enough that even the mega storage dinosaurs buy off the shelf disks because it was never worthwhile to build custom drives.
Just wanted to remind folks that dedupe is a very CPU intensive task. So when you have 24 x 2TB or 4TB drives in a shelf, getting the advertised 300 - 600TB of actual data on there will be really tough.
I know some vendors have struggled with this, but not sure of the state of dedup'ing large pools across the board.
Before worrying about an all-flash datacenter. The idea of deduplication making flash cost less is stupid, because you can dedupe to disk and realize the same savings. That's how the backup to disk products like DataDomain function.
The utility of backing up to flash is near zero, as drive performance is not a limiter and power consumption benefits would be minimal since the drives are idled when not in use. Flash will have to be get cheaper for that to make sense, and we are a long way away from that.
Chris,
I have read quite a few of your articles full of some very good points, which are often followed by your usual lot of illogical unwarranted OPINIONS. Quite predictable at this point. Can you change it up a bit please?
Let me politely educate you.
1. HDD replaced tape, and it was never because HDD was cheaper than tape. It was for many other reasons, the most fundamental being that running tape was costing businesses money by being slow and not scalable.
2. Flash does not have to be cheaper than disk to replace it therefore. Can you please re-read point 1 to understand the lesson? Hey, anyone else out there, do you want to help Chris out here?
3. If you rent a cage at a co-lo running a bunch of Hitachi HDD for 3 years in California and find yourself going over your power consumption, and having to pay overages, even though you only really occupy about a fourth of each rack in the cage, doesn't it make sense that for the next 3 years you might want to only rent 1/2 of a rack with way lower power consuming all-flash equipment, and save yourself a significant amount of money?
4. There are all-flash array companies out there that have already proven this again and again hundreds of times with hundreds of companies. But since Chris Mellor does not operate in the real world, we have to read your ridiculous unwarranted conclusions about how the all-flash data center is a pipedream still, even though the all-flash array market doubled to well over $1bil last year. The market keeps growing like that because the economics are so gray????
5. The all-flash data center was only a pip dream until all-flash companies came out with all of the Data Management functionality that all of the BS vendors have. It was a feature set void keeping these vendors from this reality, not the CAPEX argument you are trying to make here. Now that companies like Violin have launched these data management features, there is nothing stopping companies from going all-flash, and they are actively pursuing it.
Chris, come on mate. Come off of it already.
WAKE UP.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
