About this project
In the storage area, the cost of all-flash storage (SSD/NVME) is decreasing gradually, and the performance is much better than traditional HDD disk. We can see that the all-flash solution is becoming the industry trend, and all-flash based high-performance solutions are required by the market.
To provide competitive high performance all-flash storage solutions with Arm servers, there are still some big gaps we need to resolve together in the Arm server ecosystem (compared to x86). In this area, x86 has big advantages with long term investment and contribution by Intel in projects like: the next-gen Ceph solution (Crimson[1]), the ISA-L storage acceleration library[2], the DPDK[3]/SPDK[4]/PMDK[5] development kits, the DAOS[6] storage solution for HPC. But for Arm servers, we still have a lot to improve. Ceph is only functionally working on ARM64, but performance is not optimized yet (e.g, multi-core performance tuning). There are no official ARM64 releases yet for Lustre[7] and BeeGFS[8]. Besides, the ARM64 support in ISA-L is not complete either (there are some assembly optimizations specific for x86). Linaro and its members have already done some work for Ceph, Lustre, SPDK, ISA-L, etc, it can be better to have some holistic view and collaboration with a separate project to improve SDS on Arm servers.
[1] https://docs.ceph.com/en/latest/dev/crimson/crimson/
[2] https://github.com/intel/isa-l
[3] https://www.dpdk.org/
[4] https://spdk.io/
[5] https://pmem.io/pmdk/
[6] https://daos-stack.github.io/
[7] https://www.lustre.org/
[8] https://www.beegfs.io/
Goals/Objectives
The goal is to remove the gaps with x86 by collaborating together in the Arm server ecosystem, so that we can provide competitive and leading storage solutions with Arm servers.
Deliverables
a) Participate in the upstream communities of Ceph/Lustre/Rook/Mayastor/etc for ARM64 support, setup CI testing on Arm servers, and drive the official ARM64 releases;
b) Performance optimizations by leveraging key ARM64 architecture features (storage-related benchmark testing can be done for the profiling);
c) Drive community promotion by members/partners together for storage solutions on Arm servers to help customer adoptions;
Out of scope
Vendor hardware-specific optimizations should be handled by the members themselves.