Reliable data storage and access is a big challenge in AI applications and IBM ESS is a service you need to solve the problem. If you are building an AI service and you already use the IBM ecosystem for AI compute requirements, ESS can serve as an excellent software based storage solution. Integration with other platforms such as EMR/Databricks is still a challenge and should be improved.
Large scale data storage: Red Hat Ceph Storage is designed to be highly scalable and can handle large amounts of data. It's well suited for organizations that need to store and manage large amounts of data, such as backups, images, videos, and other types of multimedia content.Cloud-based deployments: Red Hat Ceph Storage can provide object storage services for cloud-based applications such as SaaS and PaaS offerings. It is well suited for organizations that are looking to build their own cloud storage infrastructure or to use it as a storage backend for their cloud-based applications.High-performance computing: Red Hat Ceph Storage can be used to provide storage for high-performance computing (HPC) applications, such as scientific simulations and other types of compute-intensive workloads. It's well suited for organizations that need to store
Highly resilient, almost every time we attempted to destroy the cluster it was able to recover from a failure. It struggled to when the nodes where down to about 30%(3 replicas on 10 nodes)
The cache tiering feature of Ceph is especially nice. We attached solid state disks and assigned them as the cache tier. Our sio benchmarks beat the our Netapp when we benchmarked it years ago (no traffic, clean disks) by a very wide margin.
Ceph effectively allows the admin to control the entire stack from top to bottom instead of being tied to any one storage vendor. The cluster can be decentralized and replicated across data centers if necessary although we didn't try that feature ourselves, it gave us some ideas for a disaster recovery solution. We really liked the idea that since we control the hardware and the software, we have infinite upgradability with off the shelf parts which is exactly what it was built for.
IBM ESS is optimized for AI and Big Data usecases while S3 is a general purpose storage solutions. EMR and Databricks have lakehouse/data warehousing solutions for distributed computing but are more optimized for just the big data pipelining solutions and not essentially for AI usecases, especially for inference, when you need to load model artifacts really quickly.
MongoDB offers better search ability compared to Red Hat Ceph Storage but it’s more optimized for large number of object while Red Hat Ceph Storage is preferred if you need to store binary data or large individual objects. To get acceptable search functionality you really need to compile Red Hat Ceph Storage with another database where the search metadata related to Red Hat Ceph Storage objects are stored.
