Rethinking All-Flash Hyperconverged Appliances

Combining all-flash and hyperconvergence isn't such a bad idea after all, Howard Marks says after seeing Atlantis Computing's appliance.

Howard Marks

May 14, 2015

4 Min Read
NetworkComputing logo in a gray background | NetworkComputing

When I wrote last year about the first demonstrations of all-flash hyperconverged infrastructure, I wasn’t all that thrilled with the concept.  It seemed to me that a hyperconverged system would use too much CPU to get the most out of its flash to be worthwhile. However, when the folks at Atlantis Computing told me that they were going to deliver a four-node hyperconverged appliance for under $100,000, I had to re-think my position.

My big mistake was worrying about “getting the most out of the flash” rather than delivering value to the user. A few years ago when flash was a rare commodity, we used it only for the most demanding workloads and needed to squeeze out every IOP we could from the few GBs of flash we could afford.

Today, however, while flash remains more expensive than disk, the gap has narrowed significantly. It’s one thing to worry about how efficiently a system uses flash that costs 10 or more times as much as spinning disk, but quite another at today’s prices. For example, Intel’s DC S3500 read-oriented SSD sells for $1/GB while the 10K RPM disks VSAN uses for its capacity layer are just under half that much.

Atlantis has taken the walk-before-you-run approach to supplying the whole hyperconverged infrastructure in an appliance. The company started with its ILIO RAM acceleration solution for VDI, then branched out into its USX software-delivered storage layer complete with data deduplication and compression.

We’ve been working with USX in my lab for months, developing a technology validation report that should be out real soon now. We ran USX through its paces with both hyperconverged volumes using server SSDs and disks and hybrid volumes across the server SSD and our old EqualLogic array.

Atlantis' HyperScale appliances combine USX with servers from Cisco, HP Lenovo or Supermicro. The Cisco, HP and Lenovo configurations use four 1U servers, while the Supermicro entry uses its four servers in 2U TwinPro2 chassis. All have two of the latest Haswell-EP Xeon E5-2680 V3 processors (12 cores at 2.5 GHz) and 256-512 GB of RAM per node.

The HyperScale appliance is a four-node cluster with the now standard simplified cluster installation and configuration wizards. Initially, there are two appliance models: the CX-12 and CX-24, which offer 12 TB and 24 TB of useable capacity respectively. By our calculations, those useable capacities assume the user’s data will reduce about 3:1, a relatively conservative estimate for virtual machine data stores, which commonly reduce by as much as 5:1. 

Figure 1:

Much as with EMC’s VSPEX BLUE, Atlantis HyperScale uses channel partners to assemble systems from Atlantis’ very explicit bill of materials. Atlantis provides the storage layer via USX, out of box experience software (a simplified installer for USX that not only installs USX, but sets up default volumes) and overall tech support, while Atlantis’ resellers can maintain their loyalty to HP or Cisco by selling their servers.

Compared to VSPEX BLUE or any other EVO:RAIL, the CX-12 offers almost twice the EVO:RAIL’s 6.7 TB of useable storage and about twice the processing power with 24 2.5 GHz cores per node vs EVO’s 12 2.1 GHz cores. That doubling of CPU horsepower should provide plenty of cycles for both data reduction and running a few more VMs.

HyperScale MSRPs start at $78,000, for the high-density SuperMicro model to a range of $125,000 to  under $200,000 for the name brands.  As usual, the vendors with the higher prices generally align with the vendors that usually discount more deeply, so the real price paid will be across a narrower range than the list prices would suggest.

While the very different software bundles make comparing EVO:RAIL and HyperScale a complicated endeavor, my back-of-the-envelope calculation shows that either product with ESXi licenses and three years of support will cost a user $140,000 to$200,000.  However, the HyperScale solution will have twice the usable storage, twice the compute horsepower, and of course a lot higher, and more predictable storage performance.

Does the HyperScale deliver every IOP the spec sheets for its SSDs promise? Frankly, I don’t care. It delivers an all-flash experience and rich data services at half the price of the market leading hybrid system. How can I complain about that being inefficient?

Atlantis Computing is a client of DeepStorage, LLC. 

About the Author

Howard Marks

Network Computing Blogger

Howard Marks</strong>&nbsp;is founder and chief scientist at Deepstorage LLC, a storage consultancy and independent test lab based in Santa Fe, N.M. and concentrating on storage and data center networking. In more than 25 years of consulting, Marks has designed and implemented storage systems, networks, management systems and Internet strategies at organizations including American Express, J.P. Morgan, Borden Foods, U.S. Tobacco, BBDO Worldwide, Foxwoods Resort Casino and the State University of New York at Purchase. The testing at DeepStorage Labs is informed by that real world experience.</p><p>He has been a frequent contributor to <em>Network Computing</em>&nbsp;and&nbsp;<em>InformationWeek</em>&nbsp;since 1999 and a speaker at industry conferences including Comnet, PC Expo, Interop and Microsoft's TechEd since 1990. He is the author of&nbsp;<em>Networking Windows</em>&nbsp;and co-author of&nbsp;<em>Windows NT Unleashed</em>&nbsp;(Sams).</p><p>He is co-host, with Ray Lucchesi of the monthly Greybeards on Storage podcast where the voices of experience discuss the latest issues in the storage world with industry leaders.&nbsp; You can find the podcast at: http://www.deepstorage.net/NEW/GBoS

SUBSCRIBE TO OUR NEWSLETTER
Stay informed! Sign up to get expert advice and insight delivered direct to your inbox

You May Also Like


More Insights