Many retail websites depend on several systems, platforms, services and applications working together. Any one of them failing can lead to lost online sales during the busy holiday season.

Murali Basavaiah, cofounder and vice president of engineering, Avi Networks

Murali Basavaiah, cofounder and vice president of engineering, Avi Networks

With the holidays around the corner, online retailers must ensure their websites can handle the onslaught. Bargain hunters swarm retailers on Black Friday and Cyber Monday. And every year many of the biggest sites struggle with performance issues, causing service disruption or even downtime. Heading into this critical season, it is important to understand the root causes of service disruption and best practices that can improve the customer experience and increase revenue.

The Impact of Site Crashes

For retailers, the holidays are the perfect storm. Not only do they need to serve the greatest number of shoppers, but shoppers determined to get the best deals. Anything keeping a purchase from completing checkout quickly will cause them to abandon the site. With shopping cart abandonment rates averaging 69% on a normal day, the stakes are higher during the holidays. Some estimates say Black Friday alone is a $3 billion online shopping day. Any outage can cost retailers significant revenue. Hourly outage costs which already average $300,000 per hour can spike into the millions on Black Friday.


The Black Friday uptime battle has now moved to third parties less accustomed to the onslaught.

The Weakest Link in the Chain

E-commerce websites are complex. They are often distributed across various data centers and clouds, integrating many third-party services, scripts and tags. A positive buyer experience depends on dozens of systems, platforms, services and applications working together. If any one slows or fails, it could cause a chain reaction affecting the entire experience. The weakest link in the chain can bring the site down.


Even in the age of cloud computing, many retailers still rely on on-premises data centers for reasons ranging from cost to compliance to control. One of the challenges of operating on-premises is that the set amount of infrastructure available. If demand for IT resources exceeds the data center’s capacity, it will cause severe performance issues or the site to crash.

It is critical for retailers to have enough capacity in their data center(s) to meet demand, or a contingency plan to scale-out or burst into another data center or cloud. Connecting multiple data centers and clouds seems like a simple process, but is complex in practice. It can take months of preparation and configuration to create a hybrid environment that maintains consistent visibility, security and control.


More and more retailers are migrating to public cloud offerings from Amazon, Microsoft and Google for elastic scale to deal with these holiday spikes. Transitioning from on-premises to the cloud involves operational and technological challenges, but making the change can provide better availability and performance during peak periods.

Application Delivery Controllers

To handle a lot of traffic, you need a lot of servers. Online retailers use application delivery controllers (also known as load balancers) to spread traffic during peak times. Counterintuitively, these very load balancers can become a bottleneck during holiday shopping. Traditional load balancers are physical appliances that reside in the data center and act as a traffic cop, processing SSL transactions and routing traffic to the best available resources. While load balancers can offload some of the burden from back-end infrastructure, they also have capacity limits. Expect outages if traffic comes in at a greater volume or rate than a load balancer can handle. Companies that use hardware load balancers need to buy enough to satisfy the surge of holiday traffic, knowing full well they will sit idle during the slower months.

Load balancers are still necessary in cloud environments, so traditional load-balancing vendors have released virtual editions of their appliances. These virtual load balancers maintain the similar architecture of their hardware counterparts, but run on virtual machines. As traffic scales in the cloud, retailers must manually configure more virtual load balancers to meet demand.

New alternatives to hardware appliance load balancers are software load balancers that run on commodity x86 infrastructure (hardware, VMs [virtual machines], or containers). These provide near real-time elasticity in the data center and across cloud environments. A software-defined or smart load balancer includes telemetry and analytics about the application. This enables predictive auto-scaling (i.e. automatically bringing more load balancers online) and is much more responsive to radical traffic fluctuations like occur on Black Friday.


Third Party Scripts, Tags and Services

In addition to infrastructure and application delivery, third-party scripts, tags and services can be a source of disruption during holiday traffic peaks. Web pages often have to load scripts and services hosted by third parties, which might include online ad servers or marketing analytics services. While a retailer’s web page may be robust, it now depends on the infrastructure and application delivery solutions of the third party as well. It’s a web of interdependencies, each of which must be able to handle the load—not just the retail site, but the third parties too.

Black Friday and Cyber Monday are too critical to have poor-performing scripts and tags degrade the customer’s experience. Many of the application monitoring and customer tracking solutions should be consolidated to reduce exposure to third-party services failing. In fact, many modern cloud infrastructure providers and software-defined load balancers provide detailed analytics that can reduce the need for third party scripts and tags too.

Historically, preparing for the holiday traffic deluge has meant buying and provisioning enough hardware for the peak, which sits idle for the rest of the year. Some sites get inundated, while for others it’s worth the cost.

Today, with elastic cloud performance, many e-commerce sites can handle their predicted load. But in the meantime, their sites have become more sophisticated and interconnected. The Black Friday uptime battle has now moved to third parties less accustomed to the onslaught. For the victors, the spoils of success are increasingly great, while others may wish they’d moved to the cloud and software-based application delivery sooner. Even so—there’s always next year.


Avi Networks provides software for load balancing, auto-scaling and security for data centers and cloud platforms.