Ceph is one of the most widely used storage systems in the world. It is a highly scalable (to exabytes) and completely distributed operation aiming to avoid single points of failure. Given the fact that Ceph replicates data for redundancy, there can be some powerful hits to its performance. However, by using Intel Optane DC SSDs as a caching tiering layer for Ceph optimized High Performance storage environments, in combination with Intel CAS or/and native CEPH cache tiering, it is possible to regain performance. ClusterVision will offer a complete end-to-end CEPH based high-performance storage solution using Optane DC SSDs for OSD caching including hardware, software and required services. Intel branded servers with JBODs and 100Gb/s Intel network as standard. Our Ceph-based Intel Optane SSD solution will be used as a single complete solution where Ceph itself is sufficient for the production environment; for those sites where HPC filesystems are required (Lustre/Beegfs etc), the CEPH based solution will act as a 2nd tier long terms backup solution.
We aim to solve our customers’ increasing challenge in managing their large HPC storage installations, particularly the disk rebuild times faced with spindle storage infrastructure. Worst still, creating proper backups is nearly impossible. HPC typically can produce >20GByte/s read/write or >>100s millions of small files in a day on a large installation. Installations have easily over 5PB worth of data in HPC, often making all-flash too expensive. With these size storage clusters, erasure-coded pools would almost be mandatory to deploy, however none of the leading HPC filesystems have had this option available. CEPH block storage supports erasure-coding however, it has been traditionally too slow for High-Performance Computing installations. Using Intel Optane DC SSDs, we can overcome a number of these obstacles to adopt CEPH as an HPC Storage environment. Combining this with NFS4.2 export capabilities, we create a solution with the at least the performance of Netapp or Dell Epsilon but without the associated price tag.
We aim to enable our customers to reliably store data for long-term backup. Ceph itself is not sufficiently performant for HPC however, with Intel Optane SSDs such as the P4800X, part of the challenge can be overcome. CEPH/Optane is very suitable for 2nd tier storage, where low latency is still an absolute requirement.
The solution would be suitable for Enterprise/Government/Academic customers using currently:
- CEPH for their production
- Customers who do not require the absolute bandwidth speeds of HPC storage, but benefit more from low latency/high IOPS
- Customers producing high quantities of daily data requiring a backup solution.