By Vik Patel
My company provides infrastructure and web hosting services for thousands of businesses, including e-commerce stores. The holiday season, which runs roughly from Black Friday to early January, is by far the busiest time for these clients. It's also the busiest time for our customer support and system administration teams.
During the holiday season, many of our customers experience huge spikes in web traffic, often many times the average. We have to organize extra staff to answer all the panicked phone calls from e-commerce merchants whose stores have slowed to a crawl or become completely unresponsive. This is particularly unfortunate because the holiday season is when e-commerce merchants make the bulk of their money. If their store falls over during this period, it can do serious harm to their business.
I advise store owners to start preparing for the holiday rush well in advance -- towards the end of summer is optimal. I'd like to share a couple of war stories from my holiday season experiences and discuss how e-commerce store owners could have avoided the worst.
The Completely Unprepared
We answer a support call: "My e-commerce store is incredibly slow and customers are emailing in complaints that they can't complete purchases." The cart is among the most resource-intensive parts of an e-commerce store. Product pages can be cached, but the shopping cart is dynamic, so code has to be executed and database queries made. We take a look at the store's infrastructure and discover that the store -- application server, web server, database and all -- runs on a single low-powered virtual private server that has been maxed out for months. It could barely cope with the site's average traffic, never mind the traffic spike it was asked to handle over the Black Friday weekend.
Analytics and resource consumption data provide all the information you need about your server's performance under load. Well in advance of the holiday season, take a close look at server and application performance and take action if it looks like your infrastructure doesn't have the headroom to cope with traffic spikes. In this specific case, I'd advise the store owner to split the database, application server and web servers onto separate servers, allowing each to be scaled independently.
Prematurely Optimized
Last year, we got a call from a new e-commerce store that was doing exceptionally well; they had great products and a fantastic marketing team. Again, their store was sluggish, especially the checkout process. This time when we looked, we saw that the store used a cluster of virtual private servers. It had multiple web servers behind a load balancer, which is great for performance and redundancy. The web servers were performing brilliantly under the heavy load -- barely breaking a sweat. Unfortunately, all those web servers were passing off requests to an underpowered database server that was struggling to keep up.
This is an example of premature optimization. Before optimizing, make sure you're optimizing something that will make a real difference in performance. Find out where the bottlenecks are, and focus your efforts there.
Lessons Learned
In both of these cases, expensive lessons were learned. First, understand how your store performs under the real-world load and make preparations for an increased load well in advance of the holiday season. Make sure you're optimizing the right thing, and understand that all the components of a system need to meet the load demand.
The end of the summer is often a quiet time for e-commerce businesses, offering the perfect opportunity for careful analysis, implementing necessary changes and upgrades, and testing them out before the doors open on Black Friday.
Vik Patel is a prolific tech entrepreneur with a passion for all things cloud. As the CEO of Detroit-based Future Hosting, Vik Patel has played a significant role in the world of hosted infrastructure since the turn of the millennium.