Client profile
Challenges
OnRule was faced with linearly increasing data center computing costs as their business scaled. They were especially concerned with the GPU compute & Data Transfer cost for their Training Datasets as they built their AI/ Machine Learning solutions using GPU servers.
The on demand pricing from AWS was suitable for them in their start up stage but was proving to be very expensive as they scaled. Also as the workloads continue to run always on it became cost prohibitive.
Solution Approach
The Opsara team was already working with the client to help them address network performance challenges as their ODC (offshore development centers) tried to connect with their end customer networks.
As part of this effort Opsara had conducted a through review resulting in gap analysis. Opsara addressed these gaps by enhancing their change management, bandwidth management practices to address latency/packet loss as issues as well as firewall configuration.
As Opsara evaluated the GPU cost escalation issue, it was evident that a customized private cloud solution would work well. The final solution included:
Designing & building a private cloud solution provided better security controls & visibility.
Optimized the number of GPUs needed by using an Acceleration on Demand solution built for AI use cases. This increased the utilization of the servers and reduced the total requirement.
Also Opsara team was able to bring to the table its expertise in provisioning the required Images with the libraries and framework such as TensorFlow.
Also implemented a Memory Caching solution to ensure low latency storage access.
The solution was built using state of the art NVDIA Tesla V100 series GPUs.
In addition Opsara worked with the client team to help with the data migration from AWS to their private cloud.
Benefits & Outcomes
- GPU server related cost reduction by 75%.
- This translated to a savings of over a million Dollars per annum at their current scale and even more savings from their future year budgets.
- Savings which the CXO team was able to reallocate for their go to market initiatives.
- Enhance greater control on their infrastructure.
- Increased end client satisfaction by providing better connectivity between the irrespective networks.
- The implementation was completed in a manner of days instead of months resulting savings realized earlier. This was accomplished using readily available proven deployment models developed by Opsara.
- The implementation was completed in a manner of days instead of months resulting savings realized earlier. This was accomplished using readily available proven deployment models developed by Opsara.
