Hazelcast Accelerates Inventory Management, Speeds Purchases

Industry

Retail

Product

Hazelcast Platform Enterprise

E-commerce sales are the growth engine for retail, returning 16% year-over-year (YoY) growth as compared to brick and mortar sales growth of 1.8% YoY. Inventory management is key for successful e-commerce. The sheer volume of e-commerce’s data and its velocity mean that architects face challenges on par with or exceeding those of the most popular social networks. A common problem is that data warehousing and inventory databases lack consistency with the data held in customer-facing e-commerce systems. Stock levels can often be minutes out of sync, and customers attempting to check out may find that the items in their cart are sold out. At best it’s a lost sale; at worst it’s a lost customer for life. Returning customers account for over 20% of retail YoY growth. That’s why major enterprises are turning to in-memory data grids – and Hazelcast in particular – to manage their inventory in real-time.

Why Most Datastores Fall Short…Why Hazelcast is Ideal…

Traditional relational databases such as Oracle or Microsoft use normalized data schemata that read-write from disk – not memory. This leads to poor application query performance resulting in unacceptable online customer experiences.

Hazelcast is a schema-less in-memory data store; it reacts much faster to query demands than typical DBMS or NoSQL solutions. Hazelcast is approximately 1,000 times faster than a DBMS.

DBMSes are expensive to scale. Typically they require individual hard drives where partitioning is not virtualized, leading to large scale misallocation of unused pockets of memory. Newer in-memory DMBSes still have massive feature sets requiring an inordinate amount of memory and server resources to scale.

Hazelcast is an easy to scale in-memory data store with all memory virtualized and redundantly maintained by software nodes. Hazelcast maximizes memory hardware utilization economies of scale.

Many NoSQL solutions such as Cassandra lack the speed required to keep up with today’s transaction volumes, as they rely on disk-based storage.

Hazelcast parallelizes and distributes processing across the nodes of the grid so that as you add more nodes throughput increases while latency decreases.

Customer Success Story

A US top ten retailer uses Hazelcast to enable real-time visibility of inventory management for their omni-channel selling platform allowing fulfillment to know that they can accurately commit to delivering a purchase within a 40ms service-level agreement (SLA). Hazelcast achieves a 2ms response time easily exceeding their SLA requirement. The platform allows a potential e-commerce customer to check inventory availability, all the way through to committing to shipping the purchase at conclusion of sale. The retailer is not only looking at the quantitative aspects of the inventory, but also qualitative–think size, color, manufacturer etc. They require real-time visibility into inventory for better order fulfillment and fewer oversell situations.

This top retailer has historically relied upon Apache Cassandra as the system of record for their platform and was experiencing 300ms of latency –well outside of their SLA. Apache Cassandra continues to be their system of record however with the introduction of Hazelcast, which is performing database caching for Cassandra, they are now experiencing only 2ms of latency. Hazelcast has been added as an operational datastore to this architecture to enable better performance. By exposing RESTful services for their e-commerce platform, and building omni-channel APIs on top, they have enabled the “check-out” step of the e-commerce platform, along with any other service, to see available inventory in real-time, using in-situ processing of the large documents. This allows for fulfillment to know that they can accurately commit to delivering the purchase by knowing where inventory is at their stores, distribution centers, drop-ship vendors, etc… Before Hazelcast, their system had 300ms of latency– missing their 40ms service-level agreement (SLA). With Hazelcast, they now have 2ms of latency and successfully exceed their business SLA.