About Service Nodes
Describes characteristics of service nodes, including boot, SDB, and login nodes. Lists other common service nodes.
Service nodes perform the functions needed to support users, administrators, and applications running on compute nodes.
Service nodes, unlike compute nodes, are generally equipped with Peripheral Component Interconnect (PCI) protocol card slots to support external devices.
Service nodes run a full-featured version of the Linux operating system. Service node kernels are configured to enable non-uniform memory access (NUMA), which minimizes traffic between sockets by using socket-local memory whenever possible.
System management tools are a combination of Linux commands and Cray system commands that are analogous to standard Linux commands but operate on more than one node. After the system is up, an administrator possessing the correct permissions can access any service node from any other service node.
Boot Node
The boot node hosts the Boot service and is controlled by Ansible. It contains the home directories on the boot RAID for local accounts, nonvolatile storage, and image roots for PE, diagnostics, and netroot. It is a tier1 node, which means it has a direct private network connection to the SMW. An administrator logs on to the boot node through the SMW console. It is booted via PXE boot and is the first node to boot.
Two boot nodes can be configured per system or per partition: one primary and one secondary for backup. The two boot nodes must be located on different blades. When the primary boot node is booted, the backup boot node also begins to boot. But the backup boot node boot process is suspended until a primary boot-node failure event is detected. For more information, see Configure Boot Node Failover.
Service Database (SDB) Node
The SDB node hosts the service database, a MySQL database that resides on a separate file system on the boot RAID. The SDB is accessible to every service node. The SDB provides a central location for storing information so that it does not need to be stored on each node. The SDB is accessible, with correct authorizations, from any service node after the system is booted.
- Global state information of compute processors. This information is used by the Application Level Placement Scheduler (ALPS), which allocates compute processing elements for compute nodes.
- System configuration tables that list and describe processor attribute and service information.
The SDB node is the second node that is started during the boot process. It is a tier1 node, which means it has a direct private network connection to the SMW.
Two SDB nodes can be configured per system or per partition, one primary and one secondary for backup. The two SDB nodes must be located on different system blades. For more information, see Configure SDB Node Failover.
Login Nodes
Users log on to a login node, which is the single point of control for applications that run on the compute nodes. Users do not log on to compute nodes.
Other Service Nodes
- MOM (machine-oriented miniserver) nodes
Note that Cray recommends treating MOM nodes (as long as they are not also SDB nodes) the same as login nodes so that they will have the same capabilities. This means that they should be assigned the same NIMS group, boot image, node group, and so forth.
- LNet router nodes
- Network gateway nodes
- DVS nodes
- Tier2 nodes for Cray Scalable Services
- DataWarp nodes
- DAL (direct-attached Lustre) nodes