Scaling cloud infrastructure
Availability of GPUs in multiple public cloud regions
The ability of Kubernetes solution deployment
Optimizing the cloud performance-to-price ratio
ReSpo.Vision faced the challenge of scaling its cloud infrastructure in a reliable and predictable manner. The company wanted to be able to accurately select the type and volume of computing resources for its workload. Given extensive use of deep learning models, GPU cards were particularly important as they are the basic processing units needed to efficiently perform data operations. Key considerations included the availability of GPUs in multiple public cloud regions, the ability to deploy the solution within managed Kubernetes clusters, and the range of solutions to ensure good value for money.
Solution and implementation
ReSpo.Vision prepared a range of detailed requirements regarding cloud infrastructure and the services the startup wanted to use.
OChK experts supported the company in building an efficient cloud strategy by clarifying and operationalizing the business goals for cloud deployment and recommending methods to easily measure success.
In order to monitor progress, the OChK team and the ReSpo.Vision team scheduled regular business and technical meetings.
Meetings focused on discussing specific tech-related challenges in the migration process. The OChK team proposed Google Cloud services relevant to the requirements, and shared their expertise on how to implement and maintain them.
Support from OChK experts allowed Respo.Vision to become fully operational on Google Cloud, seamlessly integrate new cloud services, and quickly gain awareness of best practices.
Google Kubernetes Engine
Reduced cost of infrastructure used for everyday development by 40% with a vast range of available instances enabling to choose the relevant machine size for current needs, the ability to use spot instances and a more flexible GPU machine billing method.
Reduced cost of data extraction by 30%, thanks to accurate matching of resources needed to meet the requirements of each specific module.
Reduced time-to-delivery of data to the client by 35%, with greater predictability of data processing related to resource availability in multiple regions, which reduced deployment obstruction due to lack of availability in a specific region.
Reduced DevOps team time spent on production infrastructure maintenance by 20%, thanks to greater stability of Google Kubernetes Engine (GKE) clusters and higher availability of processing resources.
"The OChK team helped to precisely translate our business objectives for Google Cloud migration into easily measurable performance metrics. This facilitated regular monitoring and enabled clear evaluation of project success. At the intersection of business and technology, we jointly agreed upon a strategy to achieve these goals by matching cloud services with the requirements. Then, in technical conversation, we were clearly presented with the advantages and disadvantages of specific products and the aspects to pay attention to when implementing them. This helped us make the right decisions. It was also important that they provided us with good practices for selected services in the migration process. The OChK team also supported us on an ongoing basis in solving any technical problems we encountered during infrastructure migration and integration of new services."
CTO, Member of the Management Board, ReSpo.Vision