Installing a compute node

A compute node is a node that runs the application of the user and instructs input/output of user data to the storage node. The compute node is connected to the storage system through the compute network.

After the connection is complete, confirm the following requirements and schedule the tasks. Then, register the information about the compute node according to Registering information about compute nodes.

The following table describes the requirements for compute nodes.

For details about how to enable ALUA and ANA, see ALUA configuration guidelines and ANA configuration guidelines. For details about other items, see the documentation for your OS.

CAUTION:

Either the ALUA or ANA must be enabled. If the ALUA or ANA is disabled, the expected I/O performance may not be achieved. In addition, the load on the internode network may increase beyond the expected level, causing delays in non-I/O processing within the VSP One SDS Block or resulting in processing failures.

For iSCSI or FC connection

Item	Requirement
OS¹	Windows Server(64bits) Red Hat Enterprise Linux(64bits) VMware ESXi
Supporting and enabling ALUA	Asymmetric logical unit access (ALUA) must be supported. In addition, ALUA must be enabled.
SCSI timeout setting	The SCSI Timeout setting value must be 120 seconds or more.²
Multipath setting (Linux Device Mapper - Multipath)	Meet all of the following: The retry count setting (no_path_retry) for the same path is 6 or more. The path polling interval (polling_interval) setting value is 30 or more. The setting value of fast_io_fail_tmo is other than off.³ The setting value of dev_loss_tmo is the OS maximum value.⁴ The failback setting is "immediate."⁵ The path_checker setting is readsector0.⁶ The hardware_handler setting is 0.⁷
MTU size of the network interface for compute network	In the case of iSCSI connection, this must be the same as the setting value of ComputeNWMTUSize in the VSSB configuration file.⁸
For details about the supported versions, contact customer support. The SCSI Timeout setting value is the time until the OS determines that a SCSI command has not responded. When the OS detects no response to the SCSI command, it performs recovery processing for the failed path. If the failed path cannot be recovered by the recovery process, the path is blocked and switched to another path. During the recovery process, I/O operations for the relevant path are stopped. Therefore, in the application layer timeout design, it is necessary to consider the recovery processing time in addition to the SCSI Timeout setting value. The default value for fast_io_fail_tmo is 5. When set to off, path switching does not occur until dev_loss_tmo seconds have elapsed when a path failure occurs, and it does not operate as expected. The maximum value depends on your distribution. See your distribution document for the maximum value. Set the failback policy ("failback") of the path group to "immediate" (immediate failback). If the failback policy is not set to "immediate", I/O will still be issued to the non-priority path after recovery from a path fault, thus requiring a manual path switch operation. If a failure occurs in a storage node of VSP One SDS Block, when a value other than readsector0 is set, paths might be wrongly blocked. For this reason, verify that you set readsector0. If a value other than 0 is set, multipath might not operate as expected depending on OS versions of the compute node. In the case of FC connection, this value does not have to be the same as the setting value of ComputeNWMTUSize because ComputeNWMTUSize is not used.

For NVMe/TCP connection

Item	Requirement
OS¹	Red Hat Enterprise Linux(64bits) VMware ESXi
NVMe timeout setting	The NVMe Timeout setting value must be 120 seconds or more.²
Multipath setting	Meet all of the following: Native NVMe Multipath must be enabled.³ The I/O policy setting for the NVMe device must be round-robin. ctrl_loss_tmo must be set to -1.⁴
For details about the supported versions, contact customer support. The NVMe Timeout setting value is the time until the OS determines that no response is returned for the NVMe command. When the OS detects no response to an NVMe command, it performs recovery processing for the failed path. If the recovery processing does not recover the failed path, the path is blocked and switched to another path. During the recovery process, I/O operations to and from the path are stopped. Therefore, you need to consider the time it takes for the OS to perform recovery processing in addition to the NVMe Timeout setting when you design timeout for the application layer. When Native NVMe Multipath is enabled, Asymmetric Namespace Access (ANA) is applied to all volumes. For details about ANA, see ANA configuration guidelines. When ctrl_loss_tmo is set to -1, path detection will be retried indefinitely if a path loss occurs.

VSP One SDS Block supports the following VMware vStorage APIs for Array Integration (VAAI) primitives.

VAAI primitive	Description	Support status
Block Zeroing	When a virtual machine is created, blocks are formatted (zeroed out). This primitive enables the storage to perform this procedure, speeding up provisioning of virtual machines. When used with the thin provisioning feature, the primitive can free the block areas that are usually reserved when a virtual machine is created, enabling efficient use of disk capacity.	Supported
Hardware Assisted Locking	If multiple virtual machines share one VMFS volume in your environment, SCSI reservation conflicts might occur and performance might degrade when Storage vMotion or a virtual machine is turned on. With this primitive, you can avoid such a problem. The primitive offers efficient locking.	Supported
Full Copy	Traditionally, VMware ESXi copy data between volumes. This primitive offloads the operation to the storage, greatly reducing the time required for virtual machine cloning and Storage vMotion processing.	Not supported