Ceph is a popular choice among cloud environments, OpenStack, Kubernetes, and other container-based platforms. It offers different interfaces to cater to various storage needs within a single cluster, eliminating the need for multiple storage solutions or specialized hardware, thereby reducing management overhead.

Ampere Computing conducted following tests on a Ceph cluster with its Arm processors.

• Block storage performance
• Object storage performance
• Ampere Arm and x86 interoperability in a single cluster
• Measuring Power consumption of Ampere and x86 servers in a cluster.
• Migrate x86 cluster to Arm with no or minimal disruption.

A snippet of Ceph block storage and power consumption on a four-node cluster with 12 OSDs is provided below.

Blog Storage Read and Write Performance

Power Consumption of a 4 Node Cluster (Q64 CPUs on Mt. Hamilton)

For more details on these tests and results, please refer our solution briefs and white papers on Ceph below.

• Ceph Solution Brief with Ampere Arm processors
• Ceph Whitepaper with Ampere Arm processors

The community version of Ceph (Quincy) was run on Ampere processors and no issues discovered in the test cases attempted. Apart from delivering good performance, we observed exceptional power savings while running Ceph. We also offered guidance for migrating existing Ceph systems on x86 processors to Ampere processors in these papers.

When designing for large scale, high performance storage, solutions architects often have to choose between performance and total cost of ownership (TCO). Specifically, the choices are between a performance requirement at a given capacity and the major operational costs associated with storage servers. Until now, system performance has been limited at the server level, not at the storage drive level, when considering performance requirements at a given capacity. The temporary workaround - not the ideal solution - is to use many smaller capacity drives. This ultimately increases the number of servers and TCO, especially when solutions architects decide to use x86 platforms as their preferred architecture.

Recently, Ampere Computing and LINBIT® have been working together to deliver a high performance, ultra-fast and resilient block storage solution that is integrated with Kubernetes to create a full-featured and production-ready solution for high-availability at a large scale. This is the ideal solution when designing for large scale, high performance storage.

To this solution, LINBIT contributes their expertise in high-availability and software-defined storage (SDS) while Ampere Computing contributes their expertise in designing and manufacturing physical processors and implementing hardware designs that maximize their potential.

A benefit of integrating LINBIT SDS with Kubernetes is that LINSTOR will not only control and manage configured block devices, it will also create and manage file system resources within Kubernetes.

See Workload Brief

LINBIT's contribution involves two essential pieces of software: LINSTOR^® and DRBD^®.

LINSTOR - The first key software component of LINBIT SDS, is open source software designed to manage block storage devices for large Linux server clusters. Its primary use is to deliver ultra-fast, resilient block storage. It also enables organizations to use commodity hardware to create high performance storage clusters.

DRBD - The second key software component of LINBIT SDS, is a distributed replicated storage system for the Linux platform, implemented as a kernel driver. When configured through LINSTOR, DRBD can be used to create larger software defined storage pools for cloud integration.

• Open source and fully supported on most major Linux distributions, including: Ubuntu Server Edition, Red Hat Enterprise Linux, SUSE Linux Enterprise Server, Debian GNU/Linux, and Oracle Linux 6.3+.
• Drivers for various platforms: Kubernetes, OpenShift, OpenStack, Apache CloudStack, ProxMox, and more.
• Manages many block storage layers: DRBD, LUKS, LVM, ZFS, VDO, NVMe, caching layers, and more.
• Separation of Data and Control Plane.
• Provides high performance, enables scale out, and low TCO.
• Hardware agnostic helps avoid vendor lock-in.