Server and rack solutions

Ceph provides both optimized server-level and rack-level solution SKUs. Hardware vendors provide both optimized server-level and rack-level solution SKUs. Validated through joint testing with IBM, these solutions offer predictable price-to-performance ratios for Ceph deployments, with a convenient modular approach to expand Ceph storage for specific workloads.

Many hardware vendors now offer both Ceph-optimized servers and rack-level solutions that are designed for distinct workload profiles. IBM works with multiple storage server vendors to test and evaluate specific cluster options for different cluster sizes and workload profiles. This work is to simplify the hardware selection process and reduce risk for organizations. IBM’s exacting methodology combines performance testing with proven guidance for a broad range of cluster capabilities and sizes.

With appropriate storage servers and rack-level solutions, IBM Storage Ceph can provide storage pools that serve various workloads from throughput-sensitive and cost and capacity-focused workloads to emerging IOPS intensive workloads.

IBM Storage Ready Nodes for IBM Storage Ceph represents an option in the IBM Storage Ceph portfolio. It comprises IBM Storage Ceph software together with a set of IBM Storage Ready Nodes as a unified offering. The Ready Nodes are IBM supported servers that are validated for IBM Storage Ceph. The IBM Storage Ready Nodes for IBM Storage Ceph offering from IBM delivers simplified IBM Storage Ceph deployment and operations, with full stack IBM support and services over the product lifecycle. For more information, see IBM Storage Ready Node for IBM Storage Ceph documentation.

IBM Storage Ready Nodes hardware recommendations for HDD cluster (cost and capacity-optimized)

Cost- and capacity-optimized solutions typically focus on higher capacity, or longer archival scenarios. Data can be either semi-structured or unstructured. Workloads include media archives, big data analytics archives, and machine image backups. Large-block sequential I/O is typical.
Table 1. Hardware recommendations for HDD cluster
Dell PowerEdge R750 X2D SATA Dell PowerEdge R760 X6D SATA HDDs
Processor Intel Xeon Silver 4314 Intel Xeon Silver 4314+ 2GHz
Number of processors 2 2
RAM 16x16GB RDIMM 16x16GB RDIMM
OS Disk 2x M.2 240GB (RAID 1) 2x M.2 480GB (RAID 1)
Data Acceleration Disk 2x 3.84 TB SSD SATA 2x 3.84 TB NVMe Read Intensive
Rack height 2U Form Factor 2U Form Factor
Width 482 mm (18.97 in.) 482.0 mm (18.98 in.)
Depth 772.11 mm (30.39 in.) 816.7 mm (32.15 in.)
Height 86.8 mm (3.41 in.) 86.8 mm (3.42 in.) | 2U
Weight 35.3 kg (77.82 lb) max 29.55 kg (65.14 lbs) / Fully pop'ed 46.3 kg (102.07 lbs)
Capacity disks 12 24
NVMe disk sizes Not applicable Not applicable
SATA disk sizes 8TB, 12TB, 16TB, 20TB 8TB, 16TB, 20TB, 24TB
Network ports 2x1GbE, 2x10GbE 2x 1GbE Management, 4x 10/25GbE, 2x 100GbE (100 Gb max)
Options / upgrades India power cords x2 Dual Port 32Gb Fibre Channel HBA, 16GB DDR5 RDIMM 4800MT/s . 2x 3.84TB NVMe Read Intensive

IBM Storage Ready Nodes Hardware recommendations for SSD cluster (throughput-optimized)

Throughput-optimized Ceph solutions are usually centered around semi-structured or unstructured data. Large-block sequential I/O is typical.
Table 2. Hardware recommendations for SSD cluster
Dell PowerEdge R760 NVMe X5D
Processor Intel Xeon Gold 6438N

Number of processors

2
RAM 16x16 GB RDIMM
OS Disk 2x M.2 480GB (RAID 1)
Data Acceleration Disk 1x 960 GB NVMeA
Rack height 2U Form Factor
Width 482 mm (18.97 in.)
Depth 772.11 mm (30.39 in.)
Height 86.8 mm (3.41 in.)
Weight 35.3 kg (77.82 lb) max
Capacity disks 24
NVMe disk sizes 3.84 TB, 7.68 TB, 15.36 TB
SATA disk sizes Not available
Network ports 2x1GbE Management,

2x 1025GbE

2x 100GbE (100 Gb max)

Options / upgrades India power cords x2
Typical rack-level solutions include the following:
Network switching
Redundant network switching interconnects the cluster and provides access to clients.
Ceph MON nodes
The Ceph monitor is a datastore for the health of the entire cluster, and contains the cluster log. Use a minimum of three monitor nodes for a cluster quorum in production.
Ceph OSD hosts
Ceph OSD hosts house the storage capacity for the cluster, with one or more OSDs running per individual storage device. OSD hosts are selected and configured differently depending on both workload optimization and the data devices installed: HDDs, SSDs, or NVMe SSDs.
IBM Storage Ceph
Many vendors provide a capacity-based subscription for IBM Storage Ceph bundled with both server and rack-level solution SKUs.
Note: For more information and assistance, contact IBM Support.

IOPS-optimized solutions

With the growing use of flash storage, organizations increasingly host IOPS-intensive workloads on Ceph storage clusters enabling emulation of high-performance public cloud solutions with private cloud storage. These workloads commonly involve structured data from MySQL-, MariaDB-, or PostgreSQL-based applications.

Table 3 lists a typical server configuration.
Table 3. Typical IOPS-optimized server configuration
Element Specification
CPU 5 physical cores (10 hyperthreads) per NVMe SSD
RAM 24 GB baseline, plus 6 GB per OSD
Networking 10-Gigabit Ethernet (GbE) per 2 OSDs
OSD media High performance enterprise NVMe SSDs, read-intensive or mixed-use
OSDs One OSD per SSD for most deployments
BlueStore WAL/DB High performance enterprise NVMe SSD, colocated on the OSD
Controller Native PCIe bus
Table 4 provides SKU solutions for IOPs-optimized Ceph workloads, by cluster size.

For more information, see Supermicro® storage products.

Table 4. Solutions SKUs for IOPS-optimized Ceph workloads, by cluster size
Vendor Small (250 TB) Medium (1 PB) Large (2 PB+)
SuperMicro 1 SSG-122B, SSG-222B, SSG-121, SSG-110P-NTR10 SSG-122B-NE316R, SSG-110P-NTR10 SSG-222B-NE3X24R

Throughput-optimized solutions

Throughput-optimized Ceph solutions are usually centered around semi-structured or unstructured data. Large-block sequential I/O is typical.

Table 5 lists a typical server configuration.
Table 5. Typical throughput-optimized server configuration
Element Specification
CPU
  • 2 physical cores (4 hyperthreads) per OSD
  • 1 physical core (2 hyperthreads) per HDD
RAM 24 GB baseline, plus 6 GB per OSD
Networking Redundant 25GE or 100GE links
OSD media 7,200 RPM Enterprise HDDs or NVMe SSDs
OSDs One per storage drive
BlueStore WAL/DB Offloaded to shared high-performance NVMe SSDs
Controller RAID HBAs are not recommended; for SAS/SATA drives a JBOD style HBA is preferred.

Several vendors provide pre-configured server and rack-level solutions for throughput-optimized Ceph workloads. Extensive testing and evaluation of servers from Supermicro and Quanta Cloud Technologies (QCT) has been conducted.

Table 6. Individual OSD servers

This table describes individual OSD servers.

Vendor Small (250 TB) Medium (1 PB) Large (2 PB+)

SuperMicro

SSG-110P-NTR10, SYS-112B-WR, SSG-122B-NE316R SSG-122B-NE316R SSG-222B-NE3X24R

QCT1

QxStor RCT-200

QxStor RCT-400

QxStor RCT-400

1QxStor Ceph Storage Open Source Edition

Table 7. Other configurable servers for throughput-optimized Ceph OSD workloads

This table describes other servers configurable for throughput-optimized Ceph OSD workloads.

Vendor Small (250 TB) Medium (1 PB) Large (2 PB+)

Dell

R670 R670, R770 R7715

Cisco

C220 M8 C240 M8 C245 M8

Lenovo

SR645 V3 SR645 V3 SR665 V3

Cost and capacity-optimized solutions

Cost- and capacity-optimized solutions typically focus on higher capacity, or longer archival scenarios. Data can be either semi-structured or unstructured. Workloads include media archives, big data analytics archives, and machine image backups. Large-block sequential I/O is typical.

Table 8 lists a typical configuration.
Table 8. Typical cost- and capacity-optimized server configuration
Element Specification
CPU 2 physical cores (4 hyperthreads) per OSD
RAM 24 GB baseline, plus 5 GB per OSD
Networking Redundant 10 GE, 25 GE, or 100 GE links
OSD media 7,200 RPM Enterprise HDDs or QLC-class SSDs
OSDs One per storage drive
BlueStore WAL/DB Colocated on the storage drive
Controller RAID HBAs are not recommended; for SAS/SATA drives a JBOD style HBA is preferred.
Table 9. Other configurable servers for cost- and capacity-optimized workloads
Vendor Small (250 TB) Medium (1 PB) Large (2 PB+)

Dell

R470 R570 R7715

Cisco

C225 M8 C245 M8 C245M8

Lenovo

SR630 V3 SR645 V3 SR665 V3