True Cost of Enterprise Flash Storage

Posted by VIOLIN SYSTEMS on Dec 29, 2014 8:50:33 AM

Enterprise Flash Storage Analysis

Today’s enterprises run on hot, up-to-date data. Active data provides near-term value to the enterprise. Data such as transactions, inventory levels, customer information, manufacturing processes and such have top and bottom line impacts on business. Data Center managers often view storage as an investment in capacity. Instead, they ought to be considering how much work data can perform. When placing storage on disk to save money, the true cost of disk storage needs to be considered. To get more work out of disk, several techniques are used that also impact the true cost of storage.

One commonly used method to get more work out of disk drives is short-stroking. This is taking advantage of the mechanics of disk drives by only using the outer tracks of the disk to read and write because the outside tracks spins faster than the inside. Aggressive short stroking will use only the outer 10% of the disk’s capacity in order to get the best possible performance from the disk. That means that disk you’ve paid $0.50/GB for really costs you $5/GB. That also means you’re now paying more than the effective price of flash storage, and still getting less work out of the disk due to inferior performance when compared to flash.

Also consider overprovisioning. There are a couple of ways to overprovision, and they are not mutually exclusive. On the disk drive model, if you have a workload that needs 5,000 IOPS of performance, and each disk drive can deliver 200 IOPS at the outer tracks, you need 25 HDDs to deliver the IOPS. Of course, if they’re 600GB HDDs, you also get 15TB of capacity which may have nothing to do with the application capacity needs. The other way to overprovision is to add a caching layer of flash. This might be a separate appliance or adding flash or even DRAM in the application server. The costs vary widely, of course, but let’s make a simplifying assumption that it will cost half of what short stroking costs, $2.50/GB.

Finally, we have the environmental costs. Spinning disk burns more energy than flash drives. The latest disk technology will be more efficient than older disk technology. Customers have seen up to 90% savings with Violin all flash arrays compared to their existing disk based arrays. If we just look at current generation disk arrays the difference might be 54% based on a power calculator on the SUN website. This works out to $4.23/GB per year less expensive than today’s disk based arrays. The cost of cooling is also less with flash than disk based systems. Using the same logic, it works out to a whopping $14.82/GB per year for cooling. And that number will be higher for older enterprise disk systems.

We started with a $0.50/GB disk drive. If we were going to add all this up, the cost of all flash storage compared to disk-based storage arrays works out to:

Short Stroking: $5/GB
Overprovisioning: $2.50/GB
Energy: $4.23/GB/year
Cooling: $14.82/GB/year

[space size="20"]

[space size="10"]

If we assume the Violin All Flash Array sells for $5/GB, that works out to $21/GB more for disk based enterprise storage over Violin All Flash Arrays. So when we say that you can get Violin All Flash Arrays for the cost of disk, we are being very generous to the disk based enterprise arrays. This quick look doesn’t include the controller or software expense. We didn’t include any of the hardware benefits from server consolidation you can achieve by getting the I/O wait times of Flash. It doesn’t include the software savings if you consolidate servers. It doesn’t include the reduced management effort for short stroking and overprovisioning. It doesn’t include the reduced floorspace available with high-density Violin All Flash Arrays compared to disk or hybrid storage arrays. We only looked at OPEX for the first year, obviously that could be extended to the total useful life of the array. Finally, it doesn’t include the incremental application benefits from sub-millisecond performance of Violin All Flash Arrays that might allow you to find new revenue opportunities like a major telecom company that discovered they were providing more minutes than they were billing for. A disk based array couldn’t solve the problem. A hybrid array couldn’t solve the problem. A Violin All Flash Array allowed them to get paid for the minutes they were already providing.

With the arrival of true enterprise data services at Violin Systems, there really isn’t a good reason to not move to All Flash Arrays from Violin. It makes sense on a CAPEX basis. It makes sense on an OPEX basis. It works on a functional basis with Violin’s Enterprise Data Services. It works on a management basis with Violin’s Symphony management.

What are you waiting for?