Kooboo Runs 5,000 Dynamic Websites on a 2-vCPU, 4 GB Server at 99.97% Success Rate

·1 views

A developer team deployed 5,000 independent Kooboo CMS websites on a single Tencent Cloud virtual private server equipped with just 2 vCPUs and 4 GB of RAM. Each site was created by importing one website package 5,000 times, giving every instance its own editable identity, content, and settings rather than sharing a single site record. A load test fired from a separate Alibaba Cloud server sent 90,000 HTTPS requests at roughly 150 requests per second over ten minutes, targeting all 5,000 hostnames. Of those requests, 89,969 returned verified, correct responses with zero HTTP errors, yielding a 99.97% first-attempt success rate and only 31 connection timeouts. Peak CPU usage reached 81.5% of the two-vCPU capacity and peak memory consumption hit approximately 2.50 GiB, with all test data and reproduction instructions published openly for independent inspection.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Five AI Terms Explained Simply: Context Windows, RAG, Fine-Tuning and More

A developer essay argues that widespread misuse of AI jargon — including conflating RAG with fine-tuning — reflects a broad gap between the vocabulary people use and their actual understanding of the tools. The piece explains five core AI concepts through plain analogies: a context window is the limited working memory an AI holds during a conversation, while RAG (Retrieval-Augmented Generation) lets a model fetch external documents before answering rather than relying solely on trained knowledge. Fine-tuning, by contrast, reshapes a model's behavior through repeated task-specific training and is not a substitute for RAG when the problem is missing information. Hallucination describes an AI generating false information with the same confident tone as accurate responses, because the model has no inherent mechanism for expressing uncertainty. The author contends that understanding these distinctions has practical cost implications, particularly when teams apply the wrong solution — such as fine-tuning a model when simply providing it with the relevant document would suffice.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Builds Always-On AI Assistant That Stores Everything Locally, No Wake Word

A developer has created an AI assistant called Karl that operates without a wake word, running continuously in the background on a user's device. Unlike mainstream voice assistants such as Siri, Alexa, or Google Assistant, Karl processes all voice data locally, with no data sent to the cloud. The assistant is designed to retain every conversation indefinitely, eliminating the need for users to repeat context across sessions. Karl can perform tasks like opening apps, making calls, and setting reminders, all while keeping user data private and off third-party servers. The project reflects a broader developer concern about AI systems that collect and store personal voice data without meaningful user control.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Australian Developer Launches Wabe Labs, Registers Brand and Domain in Week Three

Brett Anthony Ryan has officially established Wabe Labs, a solo software startup based in Australia, during the third week of his founder journey. The brand name was inspired by the word 'wabe' from Lewis Carroll's Through the Looking-Glass, chosen through an AI-assisted naming process that drew on Carroll's texts. Ryan secured the domain wabelabs.com at retail pricing and registered the business name with ASIC under a sole-trader Australian Business Number. As a sole trader rather than an incorporated entity, his personal name remains the legal contracting identity for licenses and agreements. The milestone marked a psychological shift for Ryan, who described securing the name as a concrete declaration of commitment to the venture.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Redirect Chains Let Malicious Ads Slip Past Ad Blockers

Malicious Google Ads rarely link directly to harmful domains; instead, attackers route users through a series of rapid 302 redirects that cycle through freshly registered domains faster than static blocklists can update. Because traditional blocklists only evaluate the final destination, new domains can be exploited and abandoned before they are ever flagged. Security researchers argue that behavioral signals — such as hop count, timing between redirects, and repeated stripping of referrer headers — are more reliable indicators of malicious activity than domain reputation alone. A Python-based tool has been proposed that follows redirect chains using lightweight HEAD requests, recording timing and header data at each hop to generate a dynamic risk score. The analyzer can be further customized with a YAML configuration file to set hop limits, scoring thresholds, and whitelists for known legitimate ad domains.

0 comments Read more at DEV Community