Cyber Monday, Alibaba Singles Day, Amazon Prime Day, Black Friday, Holiday Online Shopping. Let me boil it down into two words: Cyber Daze.
It is worse (and better) than you think. Cyber Monday has morphed from one day into two full months with sales that will surpass $100B according to Adobe’s reckoning. To put this number into perspective, the two months of Cyber Daze will generate enough revenue to buy every National Football League teamwith $20M left over for chips and salsa. Strange Daze indeed.
So what does $100B in 2 months have to do with our Datascape Edge-to-Cloud Blog Series and the 3D Model of Device, Data, and Data center when it comes to spending our hard-earned money online? A lot.
Consider what it would take just to capture a small fraction of these transactions with a clever online promotion. I’m thinking of a 3-hour sale on a unique product, such as the world’s greatest golf ball, with a goal of $12M in sales.
Let’s start with the technical aspects of this amazing product. It’s a colorful nighttime-illuminated golf ball featuring GPS tracking beacons that sound off and glow brightly when you touch a button on your phone or watch. If a dozen of these miracles were only $29 per transaction, we would need to execute 137,931 transactions per hour to reach our sales goal.
Given the number of pages served for search, review, and eventual purchase of this product, we would need to serve more than 8.3 million web impressions, with around 42 database requests per page or 37 million total requests during our sales window. We end up with over 10 million webserver I/Os and nearly one billion I/Os per database server just handling these requests. Managing this traffic calls for something like 12 NVMe™ servers with 2 HGST Ultrastar® SN260s, our top-of-the-line NVMe SSDs, fully maxed out. Supporting such a system, our database, advertising, and overall asset content for online marketing would easily fill numerous RU96s with HGST Ultrastar He12 hard drives. However, for this exercise we will focus on just the processing needed to handle such volumes.
Our take: it cannot be done without NVMe and scale-out architectures designed for the web, like MySQL™ and NoSQL variants. Just take a look at the technologies permeating the largest retailers and hyperscale systems. Nearly every large-scale online retailer utilizes the scale-out approach.
The reasons are numerous – CapEx, OpEx, and more importantly, traditional scale-up storage like NAS or SAN on a shared centralized database that is not designed for infinitely parallel, fail-in-place approaches, like MySQL or NoSQL databases. Furthermore, the network latency associated with a scale-up architecture is significantly higher (microseconds compared to milliseconds) which is not optimal for instant customer gratification and large quantities of orders served in parallel.
I mentioned football earlier in the blog, so let’s keep that in mind while I describe the architecture or, should I, say “the team.” NVMe SQL servers are our offensive line at the front, handling the incoming blitz of orders. Clustered to the sides for real-time analytics and recommendations we have our line coaches, also running NVMe and feeding real-time images and 4K video streams to clients looking for deeper learning and an optimized shopping experience. In the rear, we have our towering halfbacks, systems like ActiveScale™ that store massive data sets for better recommendations and inventory management. Huddled in the middle, just like a quarterback, sits our business management operations that juggle all diverse input and keep track of positions of products, time on the sale clock, etc.
To ensure that the “Wally Math” is correct here I enlisted my technical conscience, EFP3, to qualify that my “assumptions are within the realm of possibility.” Then, to make matters more fun and to invoke the holiday spirit, I took a few key statistics from our analysis to which you can apply the rhythm and rhyme of the “Twelve Days of Christmas.” Rather than go through from 1-12 and then back again, however, I just started at the top with our big goal of $12M in 3 hours!
900 Thousand/Server Reads
8 million pages served
7 thousand chat sessions
6 Servers per cluster
5 Freaked out staff
48 NVMe Drives
3– Hour Sale
And a cloud architecture for me
If it all sounds too good to be true, that’s because it is. The whole thing is entirely made up, which is a bummer because I lose way too many golf balls! Regardless, we hope you can see how the 3D world of the Datascape plays a huge role in e-commerce, from edge searching and buying to cloud processing and recommending. I hope you stay tuned as we go deeper into Digital Life and talk about more of the cool ways in which Western Digital helps make personal data thrive.
Walter (Walt) Hinton is Senior Global Director of Enterprise & Client Compute Solutions Marketing at Western Digital. He is responsible for Solutions Marketing, aligning a broad platform of storage devices to traditional and emerging data center applications. With a unique background in cloud-based services operations, product development and marketing, Walt leads a non-traditional team of Data Champions who focus on “how things work” with technical proof points, not marketing hype. Over a 25-year career in data storage, he has helped build successful startups like McDATA, ManagedStorage International and Virident Systems. He also served as Chief Strategist at StorageTek, partnering across the industry to foster the creation of the Storage Network Industry Association (SNIA).