Taiwan AI Labs Builds Scalable, Secure AI Data Management Infrastructure with Synology

Data is the foundation of AI, requiring careful planning and management during each development stage. Starting from the initial preparation stage, Taiwan AI Labs needed to store multiple petabytes of raw data. This volume of data continues to grow over time, necessitating scalable systems. Depending on the application, the data used could be sensitive or confidential, requiring stringent encryption and adherence to legal compliance.

Large parallel computing loads expose any I/O bottlenecks during the model training phase. A well-planned system with high-speed networking and consistent, sustained storage throughput is needed to maximize training performance. 

Finally, in the deployment phase, services must be flexible enough to meet final product requirements, such as high availability and continuous data security requirements. To meet these challenges, Taiwan AI Labs leveraged Synology's comprehensive portfolio of data management and protection solutions to build a solid foundation for its AI-focused infrastructure.

"Many of our new projects involve vast amounts amount of data, far beyond what we're used to working with before," remarked the teams at Taiwan AI Labs. The team designed a local object storage warehouse built on top of Synology storage systems, including all-flash units, to balance capacity, reliability, data security, and performance. 

"The bottleneck in AI training is often the I/O between the storage and compute systems, where traditional hard drives simply can't keep up with the demands," added the team. Taiwan AI Labs, after careful evaluation, tactically deployed multiple Synology FlashStations to accelerate high-I/O AI training projects.

Kubernetes (K8s) is widely adopted for its capabilities in automated deployment, scaling, and managing containerized applications. Taiwan AI Labs' environment extensively uses K8s, where they can directly manage their storage systems through the Synology CSI driver via multiple protocols (iSCSI, NFS, SMB). 

"Our goal was to eliminate single points of failure," added the team about the deployment. After additional evaluation, Taiwan AI Labs decided to pair Synology units to create Synology High Availability clusters to ensure physical disruptions could automatically be restored within minutes. 

Altogether, these storage clusters reduce the risk of service interruptions while minimizing the time needed to manage them thanks to direct integration with K8s deployments.

Taiwan AI Labs adhered to the time-tested 3-2-1 backup strategy. All data is replicated to another backup storage cluster using the built-in Snapshot Replication capability. "Snapshot Replication's incremental backup reduces the storage needed, ensuring us the ability to retain long-term backups," remarked the team about their backup setup. 

Additionally, Taiwan AI Labs stores critical data on Synology C2 Storage, completing their off-site backup requirement. "Synology C2 allows us to ensure that all data is client-side encrypted for maximum privacy," added the team. 

Building on top of reliable Synology systems, the added data protection strategies ensure that all data, whether used for training or in production environments, are safeguarded and resilient against multiple forms of data loss.