Published 5 years ago
AI for a faster website
For e-commerce, slow websites mean losing profits. But e-commerce sites need to balance the speed requirement with lots of attractive images, the transition of customers to mobile devices, and enhanced security. Machine Learning can help in all these fronts. Indeed, the use of Machine Learning to defuse security threats is nothing new. But in addition to that, we are applying Machine Learning to improve asset prioritization with HTTP/2 and to hide network latency. In the future, we also plan to use ML and other parts of our platform to enable better image compression.
The IT pyramid of needs in e-commerce
An e-commerce business necessarily includes aspects like marketing, inventories, fulfillment and so on. But the most important needs for a successful business is centered around the demands on the computing infrastructure for serving the e-commerce website. In the pyramid figure we have represented those needs and the kind of services and agencies that most typically work on setting up and managing the parts that fulfill those needs.
In the bottom layer of the pyramid we have server hardware and associated physical resources. Cloud providers operate in this layer, but we have added the qualifier "premium" to specify those which can meet certain operational demands, like unlimited server and disk space allocation, very low sharing, private intranets and operation in multiple geographical zones. Their business model is to own and maintain pools of servers and associated hardware, which they rent to customers.
The next layer up includes setting up everything needed for running a basic website. This layer can be operated by individual e-commerces with an IT team, but it is also the realm of cheap mass-market web hosting. The later focuses in packaged solutions for a specific segment: resellers of specialized merchandize, for example.
The third level from bottom to top is where most of the interesting IT needs are. The needs here are felt by e-commerce businesses with one or more of the following:
● High business volume,
● Business particularities that affect their e-commerce site or,
● High innovation rate in their particular e-commerce business.
The first point leads to more stress in the infrastructure, while the second and third lead to frequent updates in the the e-commerce software, which is likely not optimized yet for real website traffic. In those circumstances, e-commerces need better real-time information about what is happening in their infrastructure.
To satisfy needs in the third tier, e-commerces rely in consultancies with varied degrees of specialization. For example, most of these consultancies provide dedicated hardware to their e-commerce customers. Alas, most performance problems are software-related, not hardware related. Although a bit less common, there are more software and configuration-focused consultancies that can steer software optimization for their e-commerce customers. Amongst these, the services tinted in yellow in the third layer of the pyramid are the most infrequent, expensive and hardest to get.
What is ShimmerCat?
What ShimmerCat does is to enrich the top three layers of the pyramid with AI-powered solutions. ShimmerCat uses machine learning to improve the speed and ease of running of complex e-commerces. The aim of the product is to make e-commerces faster, secure and more scalable.
ShimmerCat is a disruptive solution that you can use as a web accelerator. Generally speaking, ShimmerCat is divided into two parts:
1. a set of programs, including our HTTP/2 web server software ShimmerCat installed in a so called edge server
2. a cloud analysis service where an AI engine runs
What does it do?
ShimmerCat helps improve website speed in the following ways:
- By leveraging a well-tuned, proprietary HTTP/2 implementation
- By prioritizing and pushing static assets to the browser, first while the backend is preparing the HTML, and then during the rest of the page delivery
- By stopping the most abusive automatic traffic
- By freeing resources in your application server, so that HTML responses can be produced faster
What are the expected results?
- 30%-60% loading time improvements, defined as time visitor waits to have website ready for first interaction, compared to unaccelerated setups
- Much less bot traffic, and thus enhanced website security
