Amazon Web Services China Enterprise Support Empowers Mars Hotspot to Optimize Costs and Enhance Efficiency, Accelerating Generative AI Empowerment

Maintaining timely system upgrades is one of the best practices of Amazon Web Services, which is essential for enterprises from both efficiency and security perspectives. Operational incidents caused by untimely upgrades or improper upgrade strategies are not uncommon in the industry. However, system upgrades are not easy tasks. As the system's functionalities and data volumes increase, the size of upgrade images also grows. Moreover, the increasing scale of terminal devices requiring upgrades exacerbates the insufficient bandwidth for upgrades.

OrionStar is a well-known provider of AI large models and hardware and software robot solutions in the industry, and one of the clients of Mars Hotspot. Mars Hotspot provides underlying technical operation and maintenance services for OrionStar. The upgrade of OrionStar's service robots needs to be conducted at night to ensure uninterrupted daytime operations. Therefore, Mars Hotspot must optimize its existing terminal device system update solutions without investing in new network bandwidth or other infrastructure.

As the wave of generative AI approaches, many clients aspire to leverage generative AI technology to empower their business scenarios, yet they lack the relevant technical expertise and underlying operational capabilities. In response, Mars Hotspot has taken the lead in exploring and experimenting with generative AI-related technologies, upgrading its technical service capabilities. This initiative aims to eliminate obstacles and pave the way for clients by constructing generative AI, thus enabling their business scenarios.

CTO of Mars Hotspot, Ding Guanyu, remarked, "Amazon Web Services China Region Enterprise Support are indispensable key services in our cloud infrastructure. Their expert teams often identify critical issues we encounter in our cloud deployments. More importantly, they provide broader perspectives on how we can better leverage the potential of cloud computing and meticulously guide our cloud practices. This enhances our confidence and determination in building on the cloud."

After joint research between Amazon Web Services Enterprise Support team and Mars Hotspot team, it was decided to conduct a large-scale upgrade deployment of the container service, Amazon EKS running Hunter Star through point-to-point (P2P) transmission. In the original upgrade process, terminal robots directly pulled upgrade system images from the data center. With an increasing number of terminals, the data volume of images also grew, with individual images reaching up to several tens of gigabytes. The new upgrade process incorporates P2P support, where terminals no longer directly connect to the central image but instead pull it from a node and share it with nearby nodes. This approach effectively utilizes the overall network layout and is one of the best practices for cloud deployments, effectively alleviating network congestion caused by large-scale concurrency.

Customizing large language models is a primary method for many enterprises to leverage generative AI, and RAG (Retrieval-Augmented Generation) technology is one of the fastest ways to achieve customization. By transforming relevant knowledge into vectors, RAG technology allows for better judgment of semantic similarity based on the distance between vectors, resulting in more relevant answer results.

Mars Hotspot takes the lead in exploring the prototype of generative AI technology and is one of the domestic enterprises at the forefront of the generative AI wave. Initially, an open-source vector database service was chosen, but after extensive testing, it was found to be unable to meet performance requirements and incurred high maintenance costs. Upon learning that Amazon Web Services OpenSearch now supports vector database services, Mars Hotspot opted to use this service. After multiple rounds of testing, it currently provides stable support for business services. The Amazon Web Services Enterprise Support Team recommended Amazon OpenSearch Service to Mars Hotspot, as it offers a stable vector-computing environment.

Amazon OpenSearch Service is a managed data analysis service used for interactive log analysis, real-time application monitoring, website search and more. It provides vocabulary and vector search through the kNN plugin and continuously evolves its functionality, such as adding a vector engine feature to its serverless version. These innovative features from Amazon Web Services can better assist Mars Hotspot in accelerating the implementation of generative AI for its customers.

Since migrating its EKS containerization system to Amazon Web Services in 2019, Mars Hotspot has embarked on a journey of cost optimization in the cloud with the assistance of the Amazon Web Services Enterprise Support Team. Leveraging the cost optimization technologies, tools, and best practices provided by Amazon Web Services, Mars Hotspot has employed a multifaceted approach to optimize costs in the cloud.

Firstly, Mars Hotspot began with existing cost optimization solutions, which are straightforward and yield quick results. In terms of pricing models, Mars Hotspot selects the most suitable model based on varying workload demands. Amazon Web Services offers on-demand, reserved, and the newly introduced Savings Plans pricing models in China region this year. The latter two models can significantly reduce instance costs compared to on-demand instances. For example, Savings Plans can save up to 84% on costs for services like Amazon EC2, Amazon Fargate, and Amazon Lambda. Regarding storage, Mars Hotspot strategically utilizes storage services with different prices and performance levels to improve overall costs. For example, it stores archive data with low access frequency requirements in a lower-cost cold storage tier or employs Amazon EBS's next-generation general-purpose volume GP3, saving approximately 20% per GB compared to the previous GP2 volumes.  

Secondly, Mars Hotspot focuses on cost savings through architectural optimizations based on cloud containerization best practices. After in-depth communication and research with Mars Hotspot, the Amazon Web Services Enterprise Support team proposed reusing nodes to reduce the number of EKS clusters, achieving cost savings without affecting business functionalities.

With the assistance of Amazon Web Services Enterprise Support Team, Mars Hotspot has acquired advanced capabilities in EKS containerization operations. Mars Hotspot is also a member of the Amazon Web Services Partner Network (APN) system, holding MSP certification, as well as certifications in migration, DevOps, data analytics, and EMR big data fields. Mars Hotspot aided OrionStar in achieving automated EKS upgrade deployments using scripts. Simultaneously, it addressed the slow EKS system upgrade issue without additional hardware investments, reducing upgrade time by 50% and ensuring uninterrupted business functionalities of OrionStar's terminal service robots.

Mars Hotspot's operational and innovative capabilities provide reliable support for the development of OrionStar's robot business. As of December 2023, Hunter Star's service robots have been deployed in over 60 countries and regions worldwide, serving nearly 550 million people, with a total shipment volume exceeding 60,000 units.

Through a multifaceted approach to cloud cost optimization, Mars Hotspot has emerged as one of the best practices in this field, achieving a remarkable optimization rate of up to 60% compared to initial costs. Beyond system operations, Mars Hotspot has made significant progress in the field of generative AI. For instance, in its cloud cost management platform, SavingsNow, Mars Hotspot launched the BIChat feature, enabling the generation of charts and reports, significantly enhancing the efficiency of cloud cost optimization.

Ding Guanyu, CTO of Juyun Technology, stated, "With the increasing focus on cloud costs, how to use the cloud more economically is a topic that enterprises should pay more attention to. Through the Amazon Web Services Enterprise Support Service, we have truly achieved cost reduction and efficiency improvement, allowing us to devote more energy to exploring core business and cutting-edge technologies such as generative AI. Recently, we have applied large language model capabilities to our self-developed cloud cost management platform, SavingsNow, and introduced the BIChat feature, which supports generating charts and reports using natural language. This has reduced the time required for producing cloud cost analysis reports from 2-3 days to just 30 minutes, completely revolutionizing the efficiency of enterprise cloud cost optimization implementation."

Mars Hotspot is a multi-cloud management service provider, dedicated to its mission of "achieving customer success through expertise." Leveraging software-defined services, it assists enterprises in their digital transformation journey, enabling rapid achievement of unified management, elastic expansion, performance enhancement, and cost optimization.