What a year it’s been for FlashBlade! Since its unveiling at //Accelerate 2016 a year ago, and its general availability six months ago, FlashBlade has been deployed into every continent (besides Antarctica), with large-scale deployments in industries as varied as SaaS, Consumer Electronics, Social Media, Finance, Service Provider, Healthcare, Genomics, Gaming, Manufacturing, and Consumer Internet.
In this blog, I’m excited to introduce the next generation of “big” and “fast” for FlashBlade. If you missed our other announcements, I encourage you to browse these blogs for more details.
Cloud-Era Flash: “The Year of Software” Launch Blog Series
Powering one of the world’s most powerful AI supercomputers. Accelerating simulations of the world’s fastest racing cars. Enabling ultra-fast database restores for a SaaS leader. Then of course, there are trains, planes, cars, and rockets. It’s been a busy year for the FlashBlade team.
FlashBlade TM is the industry’s first cloud-era flash purpose-built for modern analytics. The word “purpose-built” is key for two reasons. First, every other system for big data is retrofitted using decades-old software and hardware building blocks. Second, when a purpose-built system is engineered effectively, it can have revolutionary impact. For a world leading market intelligence and information services company in Europe, data is its lifeline. Prior to FlashBlade, the customer deployed 20 racks of disk, along with a physical building to host all the racks, power and cooling, and everything else that’s needed for the infrastructure. With a single 4U FlashBlade, they replaced all 20 racks without compromising on capacity and dramatically increasing performance, all within space of a microwave. Now, THAT is revolutionary.
However, customers pushing the boundaries of what’s possible, those building the fastest AI supercomputers and web-scale applications, want more from their data platform. They want faster, bigger, yet still simple. So we’re delighted to announce a new FlashBlade, one that can scale to 75 blades with a single IP and namespace.
5x faster. 5x bigger. Still simple. All within a year of unveiling the first 15-blade FlashBlade! Powered by the new Purity for FlashBlade 2, all the resources in 75 blades, flash chips, CPU cores, FPGAs, and more, are seamlessly managed and aggregated with unprecedented performance and simplicity. This feature will be available in Q4 2017 and will deliver 600 CPU cores. 225 FPGAs. 1.9 TB of DRAM. 8.0 PB of effective capacity. ½ rack. While these eye-popping stats are breaking new grounds, what is more amazing is the delivered performance. Performance scales perfectly linear with more blades, up to 75 GB/s read throughput, 25 GB/s write throughput, and 7.5M IOPS.
Massively distributed computing that delivers linear scaling is a very difficult computer science problem. Computer scientists have struggled with parallel processing and programming as it is exponentially more difficult than serial processing. But when a solution is created, like Purity for FlashBlade, it can yield incredible results for customers who have large scale problems to solve.
We live in a special time in human history. The World Economic Forum has called this era the 4th Industrial Revolution, driven by innovation in artificial intelligence, big data, and internet of things. New algorithms, like Spark and machine learning, coupled with modern parallel processors, like CPUs and GPUs, are fueling its rapid rise. Just in two years, compute required to train deep neural networks jumped 15x and compute delivered by GPUs increased by 10x.
While algorithms and compute are massively parallel, legacy disk-based storage or retrofit SSD-based storage are largely based on serial technologies- NFS software stack, SAS and SATA protocols are decades old. The performance gap between compute and storage will continue to grow.
If data is the new currency for the fourth industrial revolution, why is data still stuck in the serial era?
FlashBlade is built from day one to be massively parallel, enabling us to deliver 5x performance jump in one year. This is truly the power of purpose-built architecture that’s engineered from the ground up for Spark, Hadoop, machine learning, and other modern analytics workloads.
We hope you are excited as we are. It’s been an amazing year, but Pure engineers are not done yet.
We’d love to hear from you. Please use the comment box below on how 75 blade-scale in ½ rack can change what’s possible for your organization.