Over the last few years all-flash memory storage array providers have made many claims about the cost savings that all-flash arrays provide. Examples include lower operating costs, faster performance, smaller data center footprints and acceleration of server virtualization and consolidation efforts. Now those claims are turning into proof points. In a recent conversation with Cloudmark’s VP of Operations, Ryan White, he explained how Cloudmark implemented the Nimbus Data Gemini all-flash array and was able to claim a half million dollars in annual savings as the prize.
Almost any time a new technology comes to market, the hype around it often overshadows the technology itself. In this respect, all-flash memory arrays have been no different. In fact, as recently as a few years ago I was at a trade show where a mouthpiece for a tape provider proclaimed that within the next ten years hard disk drives (HDDs) not tape would be dead thanks to flash memory. Not sure I ever agreed with that statement but it illustrated the type of hype that flash was generating in the industry.
Thankfully sounder minds have prevailed since then but interest in flash memory persisted. As organizations explored and understood this technology, they found that it still had shortcomings. While it certainly outperformed all comparable storage technologies, it did not have the same management and support features that other HDD-based storage arrays offered. As such, the adoption of flash memory has not been as fast as some anticipated.
That is beginning to change as evidenced by my recent conversation with Cloudmark’s VP of Operations, Ryan White. This is an individual who had already virtualized 75 percent of Cloudmark’s servers. But in order to go the final mile and achieve 100 percent virtualization, he had to virtualize Cloudmark’s most performance intensive applications.
Unlike other organizations that can sometimes afford to wait, White had a carrot dangling out in front of him. If he virtualized these servers (which amounted to about three (3) racks of servers in Cloudmark’s San Francisco data center,) Cloudmark could close this data center, consolidate its operations into its San Jose data center and realize $500,000 in annual savings. That would put a little spring in any company’s step and it certainly helped to motivate Cloudmark to get this project done.
Like many large companies do, it first evaluated storage solutions from providers such as EMC and NetApp. However they did not meet his requirements. Their larger storage system models cost too much, took up too much floor space, and consumed too much power. Their smaller models did not provide enough performance to meet Cloudmark’s anticipated I/O requirements.
This prompted White to turn his attention to the new generation of flash memory storage arrays. These systems offered lower price points, fit into existing data center racks, consumed a fraction of the power and, most importantly, provided the performance Cloudmark needed.
To select the right one, he brought in models from Nimbus Data, Tegile, and WhipTail (now Cisco), and ran head-to-head comparisons against them. White says, “Nimbus Data ended up coming out on top because of performance, rack space savings and ROI to be achieved. During testing we achieved 250,000 IOPS per 8Gb FC port which would easily meet the IOPS that our existing physical systems generated.”
As White prepared to introduce Nimbus Data into Cloudmark’s infrastructure, he recognized that he had the opportunity to achieve even further cost savings. Cloudmark hosted VMware vSphere 5.1 on HP ProLiant DL360 Gen8 servers that each included two Emulex Converged Network Adapters (CNAs). Using these two CNA adapters and their 10Gb Ethernet connectivity, Cloudmark could connect its servers to a converged Brocade VDX 6730 Switch using Ethernet. The Brocade VDX 6730 could then connect to either Fibre Channel (FC) SAN or Ethernet attached storage arrays.
On the surface, this sounded great to White. Using Ethernet instead of FC would eliminate the need for Cloudmark to acquire more hardware and rack space while also reducing its power consumption, capital investment and management requirements. These benefits aligned perfectly with Cloudmark’s cost saving initiatives.
However before moving forward with this FCoE solution, White had to determine if this configuration of 10Gb Ethernet-attached CNAs, two Brocade VDX 6730s and an 8Gb FC-attached Nimbus Data Gemini all-flash array could deliver sufficient performance to meet Cloudmark’s needs. White says, “We rigorously benchmarked this configuration before we put it into service. Not only was the converged infrastructure able to support the IOPS that we needed but the FCoE solution seamlessly integrated into our environment and worked right out of the box.”
My conversation with Cloudmark’s VP of Operations, Ryan White, illustrated just how far the adoption of flash memory storage arrays has progressed in the last few years. While flash memory may never be the death knell for HDDs, all-flash arrays are rapidly finding new homes in enterprise data centers that want to finish their data center virtualization initiatives and are displacing HDD-based arrays to do it as Cloudmark’s use of the Nimbus Data Gemini all-flash array illustrates. As this occurs, companies like Cloudmark are finding the performance so much better and the footprint so much smaller, they are also able to use more cost-effective storage networking technologies, reduce operational costs and even close entire data centers.