IT hardware for a Ceph storage cluster

Detailed performance requirements for the subject of procurement are defined in the catalog of requirements. Any deviation from the characteristics of the specified minimum requirements is only permitted in a positive sense (benefit to the contracting authority), e.g., through additional function(s) or by exceeding the required characteristic(s). In case of doubt, the bidder must at least demonstrate equivalence.

In the event of a negative deviation, the contracting authority reserves the right to seek subsequent clarification and, within its discretion, examine the possibility of excluding the tender. If equivalent products are expressly permitted, the mandatory minimum requirements must still be complied with in any case.

3.1 Storage Nodes

Each storage node must be powerful enough for a full population of HDDs and meet the following requirements:

3.1.1 Chassis & Form Factor

• 19" rack mount, maximum 2 U per node
• Suitable for 24/7 data center operation
• Redundant, hot-swappable power supply units (at least 2 x 100% capacity)
• Redundant cooling with N+1 configuration (sufficient cooling capacity in case of fan failure)
• Integrated remote management (iLO/BMC/IPMI) with dedicated management IP 3.1.2 Processor and Memory
• One CPU (single-CPU server) with at least 48 physical CPU cores (comparable to AMD EPYC Genoa 9454P, Intel Xeon Scalable 3rd Gen. or newer, or equivalent. Older generations do not fulfill the minimum requirements for the scope of services. In case of doubt, currency must be verified by the bidder.)
• At least 192 GB DDR5 ECC-RDIMM (6400 MT/s), distributed across at least six 32 GB modules
• Built-in, dedicated storage controllers for NVMe and HDDs 3.1.3 Storage
• 12 × 20 TB SAS Enterprise HDD (3.5"), at least 24 hot-swap bays
• Upon delivery, at least 12 bays populated with HDDs, and at least 12 bays free
• Expandability to at least 24 HDDs without replacing the base platform
• 4 × at least 3.84 TB Enterprise TLC NVMe SSD (M.2 or U.2)
• 2 × at least 960 GB M.2 TLC NVMe as redundant boot storage (e.g. for Ceph OSD boot image)
• HBA/JBOD/pass-through controller for direct access to the Ceph storage media
• No traditional hardware RAID for Ceph OSDs - all drives must be individually recognized and addressable by the operating system 3.1.4 Network
• 2 × 10 Gb for client/public network
• 2 × 100 GbE for dedicated Ceph cluster network 3.1.5 Operating System and Ceph Compatibility
• Hardware must be compatible with current stable Ceph versions and enterprise Linux distributions (e.g., RHEL, Debian, Ubuntu LTS)
• Proof of compatibility must be included with the tender

3.2 Ceph Cluster

• The hardware must be suitable for stable, scalable, and reliable Ceph operations
• The HDDs and NVMe must be individually recognized and addressable by the operating system (no RAID abstraction)
• Nodes are to be distributed as evenly as possible across three racks (e.g., 5 nodes per rack for 15 nodes).
• In the future, the three racks are expected to form separate physical failure domains

3.3 Cluster Network

3.3.1 Requirements for the Nodes

Each node must support 2 x 100 GbE, redundantly connected to two separate switches.

3.3.2 Switches

• Two 100 Gbps switches
• At least 24 × 100 Gbps ports per switch (at least QSFP28)
• At least 20% free ports after initial installation
• Non-blocking (full bandwidth with full port traffic)
• Redundant power supplies (N+1)
• Redundant fans (N+1)
• Jumbo frames ≥ 9,000 bytes
• VLAN, LACP or technically equivalent link aggregation
• IPv4/IPv6, QoS, LLDP, SNMP, Syslog, NTP, SSH, web and CLI management
• Failure of one switch must not completely interrupt the 100 Gbps cluster connection of the nodes (redundancy ensured via second switch) 3.3.3 Cabling All transceivers (e.g., SFP28, QSFP28 etc.), DAC (Direct Attach Copper), AOC (Active Optical Cables), and fiber-optic connections required for operation

3.4 Client Network

• Each node must provide 2 × 10 GbE for client/public communication. The co…