This topic mainly describes the hardware configuration requirements for servers, including the operating system, BIOS settings, disk mounting, network adapter settings, software requirements, and more.
Prepare the operating system
Operating systems for OAT/OCP
You can deploy OAT/OCP on the operating systems listed in the following table.
Operating system |
Supported Versions |
Server Type |
|---|---|---|
| Rocky Linux | 9 | x86_64/ARM aarch64 |
| Alibaba Cloud Linux | 2、3 | x86_64/ARM aarch64 |
| Anolis OS | 8 | x86_64/ARM aarch64 |
| Kylin OS | V10、V11 | x86_64/ARM aarch64 |
| UnionTech UOS | V20 | x86_64/ARM aarch64 |
| NFSChina | 4.0 | x86_64/ARM aarch64 |
| CentOS/RHEL(Red Hat Enterprise Linux) | 7、8、9 | x86_64/ARM aarch64 |
| openSUSE | 12SP5 | x86_64/ARM aarch64 |
| Debian | 12 | x86_64/ARM aarch64 |
| openEuler | 20.03 LTS、22.03 LTS | x86_64/ARM aarch64 |
| LinxOS | V6.0.99、V6.0.100 | x86_64/ARM aarch64 |
| Ubuntu | 22.04 LTS、24.04 LTS | x86_64/ARM aarch64 |
Operating systems for OceanBase AI Database
OceanBase AI Database consists of two components: OBServer and LogService. You can deploy OBServer and LogService on the Linux operating systems listed in the following table.
Linux operating system |
Version |
Server Architecture |
|---|---|---|
| Rocky Linux | 9 | x86_64 (including Hygon), ARM_64 (Kunpeng, Phytium) |
| Alibaba Cloud Linux | 2、3 | x86_64 (including Hygon), ARM_64 (Kunpeng, Phytium) |
| Anolis OS | 8 | x86_64 (including Hygon), ARM_64 (Kunpeng, Phytium) |
| KylinOS | V10、V11 | x86_64 (including Hygon), ARM_64 (Kunpeng, Phytium) |
| UnionTech UOS | V20 | x86_64 (including Hygon), ARM_64 (Kunpeng, Phytium) |
| NFSChina | 4.0 | x86_64 (including Hygon), ARM_64 (Kunpeng, Phytium) |
| Inspur Kos | 5.8 | x86_64 (including Hygon), ARM_64 (Kunpeng, Phytium) |
| CentOS / Red Hat Enterprise Linux | 7、8、9 | x86_64 (including Hygon), ARM_64 (Kunpeng, Phytium) |
| SUSE Enterprise Linux | 12SP5 | x86_64 (including Hygon) |
| Debian | 12 | x86_64 (including Hygon) |
| openEuler | 20.03 LTS、22.03 LTS | x86_64 (including Hygon), ARM_64 (Kunpeng, Phytium) |
| LinxOS | V6.0.99、V6.0.100 | x86_64 (including Hygon), ARM_64 (Kunpeng, Phytium) |
| Ubuntu | 22.04 LTS、24.04 LTS | x86_64 (including Hygon) |
Note
- Before you use an operating system, configure the network and install a package manager such as YUM and Zypper first.
- The server used to deploy the OceanBase AI Database cluster must be in little-endian mode.
- If you use the Hygon 7490 chip, you must install the operating system that contains Patch 4 as recommended by the Hygon server vendor.
- The Linux kernel version must be 4.19 or later.
Configure BIOS for OceanBase AI Database
Special settings
In the Intel x86 architecture, we recommend that you:
Run the following command to set
vm.swappinessto0in the/etc/sysctl.confconfiguration file:[root@xxx /] $vi /etc/sysctl.conf vm.swappiness = 0Apply the values from the
sysctl.confconfiguration file:[root@xxx /] $sysctl -p
In the AMD or ARM architecture, we recommend that you enable Numa.
In the ARM or Hygon architecture, we recommend that you run the following command to set the values of the
kernel.numa_balancing,vm.zone_reclaim_mode, andvm.swappinessparameters to0in the/etc/sysctl.confconfiguration file:[root@xxx /] $vi /etc/sysctl.conf kernel.numa_balancing = 0 vm.zone_reclaim_mode = 0 vm.swappiness = 0Apply the values from the
sysctl.confconfiguration file:[root@xxx /] $sysctl -p
BIOS options to be disabled
Cstate
Pstate
EIST
Power saving
SMMU (ARM architecture)
BIOS options to be configured
Automatic Power on After Power Loss: Always on
Intel Virtualization Technology: Enable in virtualized scenarios and disable in non-virtualized scenarios
Hyper-threading: Enable
Hardware prefetcher: Disable
VT-d: Enable in virtualized scenarios and disable in non-virtualized scenarios
SR-IOV: Enable in virtualized scenarios and disable in non-virtualized scenarios
SMMU/IOMMU (IOMMU for the AMD architecture and SMMU for the ARM architecture): Enable in virtualized scenarios and disable in non-virtualized scenarios
Turbo Mode: Disable
Energy performance: Enable the maximum performance
Note
The methods for modifying the BIOS settings vary among different servers. For specific operation steps, see the operation instructions of your server.
Disk mounting
Notice
If a file in a mounting point directory is larger than 16 TB in size, only the XFS type is supported.
The following table lists the disk mounting point requirements for an OCP server.
Mounting pointSizeDescriptionDisk format/home 100 GB to 300 GB The operation log disk of the components. Ext4 or XFS is recommended. /data/log1 3 to 4 times the memory size The log disk of the OCP MetaDB. Ext4 or XFS is recommended. /data/1 Depends on the size of data to be stored The data disk of the OCP MetaDB. Ext4 or XFS is recommended. /docker 200 GB to 500 GB The docker root directory. Ext4 or XFS is recommended. The following table lists the disk mounting point requirements for an OBServer node.
Mounting pointSizeDescriptionFile system format/home 100 GB to 300 GB The operation log disk of the observer process. Ext4 or XFS is recommended. /data/1 Depends on the size of data to be stored The data disk of the observer process. Ext4 or XFS is recommended. Note
We recommend that you allocate at least 50 GB for a root directory. If you use Logical Volume Manager (LVM), we recommend that you specify the stripe parameters during the creation of directories. Here is an example:
lvcreate -n data -L 3000G obvg --stripes=3 --stripesize=128The following table lists the disk mounting point requirements for a LogService node.
Mount pointSizePurposeFile system format/home 100 GB to 300 GB The operation log disk of the logservice process ext4 or xfs recommended /data/log1 100 GB Log disk for the logservice process ext4 or xfs recommended
NIC settings
We recommend that you configure two 10 Gbit/s NICs:
- Name the bond interface bond0. You can select either mode1 or mode4 for bonding, but mode4 is recommended. For mode4, the switch must be configured with the 802.3ad protocol.
- We recommend that you name the NICs as eth0 and eth1.
Recommendations for SSD I/O schedulers
- The default I/O scheduler for NVMe SSDs is
none, so no adjustment is required. - We recommend setting the I/O scheduler for SATA SSDs to
none.