It was just last week that Coho Data engineering team unveiled their latest creation: FlashBlade. FlashBlade is a blade-based flash storage array that addresses a problem that users have openly discussed over the past few years which is that the performance density of the PCle flash demands a completely fresh set of architectures for storage systems. In 2013, the Coho Data engineering team decided to pass on building a similar platform for FlashBlade because the main focus was developing the initial version of Coho’s storage software.
This unsurprisingly left little effort to build the densest, network integrated hardware platform that is needed today. Back in 2012 Coho saw these problems very clearly as the team started working with PCle flash. One needed a lot of CPU because the cards themselves were so fast that a single device could easily saturate a 10Gb NIC. This led Coho Data to build systems that could scale compute and integrate storage and connectivity together. The only remaining issue was density as these flash devices were incredible yet needed to be paired with compute and network to really scale the way it needed to.
The challenge for Coho Data was how to pack all of these things together as tightly as possible into a single physical form. Working with Intel and other boutique design firms, Coho began to explore the idea of integrating Intel’s Seacliff Trail 10Gb top of rack switch architecture as a bsis for a backplane of a blade-based storage server. This prototype system was christened SwitchStore with the idea of creating a completely converged top of rack switch and storage system.
However with the design in place Coho Data decided not to go ahead with SwitchStore due in part to a variety of issues. The first is that building a system based on proprietary hardware assuredly involves longer cycle times, higher costs and bigger challenges in validation and quality assurance.
Building their own hardware meant that Coho Data would be slower to hit the market on a less established and reliable platform all the while falling behind to all the other new technologies while they are still refining theirs. The second issue is that scaling out is a network problem.
In order for the array to consolidate the enormous amounts of storage performance into a single rack the loaded chassis has to be kept busy which would involve compute from multiple other racks. The resulted in a need to draw high-rate storage traffic across the core of the network. If Coho was to scale-out storage performance, they needed to do it in a way that is flexible enough to appropriately address the broader topology/network efficiency of your datacenter.
Finally scale-out essentially is expected to resolve a business problem which means it’s all about flexibility. Customers want the flexibility to maximize efficiency similar to those offered by scalable public cloud-based storage systems. Proprietary hardware would mean that a customer pays more for a bespoke, lower-volume device that is locked onto a backplane/topology that is slow to evolve.
FlashBlade is being marketed as a scale-out storage product and while it is quite an expensive proprietary piece of enterprise storage hardware that has a dedicated software team that writes enterprise code for 3 different processor architectures in Xeon, ARM, and FPGA. Only time will tell if Coho Data’s FlashBlade will solve all the issues limiting its predecessor SwithStore but for now the company remains optimistic it has finally cracked the code.
Here at ASAP Memory, we have a dedicated and expansive array of Coho Data Inc. products. We are your one-stop shop and go destination for a simplified sourcing solution. ASAP will ensure that our consumers’ needs are addressed in the most expeditious and transparent manner all the while offering cost-effective component solutions therefore improving our your negotiation power and profit margins. If you are interested in a quote, please don’t hesitate to contact our friendly sales staff at www.asap-memory.com or call us at 1-714-705-4780.