Posted by: Gagan Mehra
Analysis, BigMemory, In-memory data management, Retail & E-commerce
With the holiday season around the corner, online retailers are looking forward to big spikes in sales, but their CIOs are hoping that their sites will be able to handle the demand. The smartest CIOs began planning for this many months ago by implementing In-Memory Data Management (IMDM) solutions with low, predictable latencies—even while coping with enormous customer traffic.
At Terracotta, we’re seeing online retailers adopt our IMDM solution, BigMemory, as a core strategy for getting real-time performance at scale for critical components of online retailing. In particular, we’re seeing e-retailers put more and more of the following in BigMemory:
1. Customer Information: Retailers must store customer account details, preferences, order histories, wish lists, and much more. Loading all of this information in memory at the time of customer authentication improves performance and reduces database load, as requests for customer information will read from memory instead of disk.
2. Product Catalog: Since product catalogs are relatively static, most sites use caching to keep the catalog in memory. But caching has two big constraints: (1) product catalogs are typically too big to fit entirely in the cache, so only the most frequently accessed items are quickly accessible; and (2) if the site runs on Java, then the catalog can fill the JVM heap, leaving little memory for other parts of the application and causing unpredictable garbage collection pauses. An IMDM solution overcomes both of these constraints by allowing your Java application to use all available RAM in distributed servers. You get to keep not only the full catalog in memory, but also prices and promotions.
3. Reviews and Recommendations: Most online retailers use third-party services for product reviews and recommendations, which can be the source of significant latency for customers. With an IMDM solution, you can load product reviews and recommendations into memory when the server starts and simply refresh them periodically.
4. Search: Without effective, fast search, a site risks losing customers. IMDM makes search faster and more scalable by keeping all search indexes in memory. This requires your search index process to interact directly with your IMDM solution to update in-memory indexes.
5. Fraud management: Online fraud detection is often overlooked as a source of profit by retail CIOs, who fear that fraud detection alogrithms will add so much delay to a checkout that customers will abandon purchases or that systems will timeout. But just as BigMemory has helped major credit card and financial companies execute complex fraud detection in under 250ms, it can scale to thousands of transactions per second to minimize fraud loss. One of Terracotta’s largest customers has seen a 4000% improvement in fraud detection performance after deploying BigMemory, resulting in a bottom-line bump of starting at tens of millions of dollars annually.
Know of other ways in-memory data management will help retailers during this (and future) holiday seasons? Post a comment, or drop me a line at email@example.com.