NEMO/Hardware

System Architecture

The bwForCluster NEMO is a high-performance compute resource with high speed interconnect. It is intended for compute activities related to research in for researchers from the fields Neuroscience, Elementary Particle Physics and Microsystems Engineering (NEMO).

Figure: bwForCluster NEMO Schematic

Operating System and Software

• Operating System: CentOS Linux 7 (similar to RHEL 7)
• Queuing System: MOAB / Torque (see Batch Jobs for help)
• (Scientific) Libraries and Software: Environment Modules

Compute Nodes

For researchers from the scientific fields Neuroscience, Elementary Particle Physics and Microsystems Engineering the bwFor Cluster NEMO offers 748 compute nodes plus several special purpose nodes for login, interactive jobs, etc.

Compute node specification:

	Compute Nodes	Interactive Nodes	Memory Nodes	Special Purpose Nodes
Quantity	772	6	4 / 4	4
Processors	2 x Intel Xeon E5-2630v4 (Broadwell)			1 x Intel Xeon Phi 7210 Knights Landing (KNL)
Processor Frequency (GHz)	2,2			1,3
Number of Cores per Node	20			64
Working Memory DDR4 (GB)	128		256 / 512	16 GB MCDRAM + 96 GB DDR4
Local Disk (GB)	240 (SSD)
Interconnect	Omni-Path 100

Special Purpose Nodes

Besides the classical compute node several nodes serve as login and preprocessing nodes, nodes for interactive jobs and nodes for creating virtual environments providing a virtual service environment.

Storage Architecture

The bwForCluster NEMO consists of two separate storage systems, one for the user's home directory $HOME and one serving as a work space. The home directory is limited in space and parallel access but offers snapshots of your files and Backup. The work space is a parallel file system which offers fast and parallel file access and a bigger capacity than the home directory. This storage is based on BeeGFS and can be accessed parallel from many nodes. Additionally, each compute node provides high-speed temporary storage (SSD) on the node-local solid state disk via the $TMPDIR environment variable.

	$HOME	Work Space	$TMPDIR
Visibility	global (GbE)	global (Omni-Path)	node local
Lifetime	permanent	work space lifetime (max. 100 days, extension possible)	batch job walltime
Capacity	45 TB	768 TB	200 GB per node
Quotas	100 GB per user	20 TB / 1 Million files per user	none
Backup	snapshots + tape backup	no	no

 global             : all nodes access the same file system
 local              : each node has its own file system
 permanent          : files are stored permanently
 batch job walltime : files are removed at end of the batch job

$HOME

Home directories are meant for permanent file storage of files that are keep being used like source codes, configuration files, executable programs etc.; the content of home directories will be backed up on a regular basis. The files in $HOME are stored on a Isilon OneFS and provided via NFS to all nodes.

 NOTE:
 Compute jobs on nodes must not write temporary data to $HOME.
 Instead they should use the local $TMPDIR directory for I/O-heavy use cases
 and work spaces for less I/O intense multinode-jobs.

Workspaces

Workspaces can be generated through the workspace tools. This will generate a directory on the parallel storage with a limited lifetime. When this lifetime is reached the workspace will be deleted automatically after a grace period. Workspaces can be extended to prevent deletion. You can create reminders and calendar entries to prevent accidental removal.

To create a workspace you'll need to supply a name for your workspace area and a lifetime in days. For more information read the corresponding help, e.g: ws_allocate -h.

Defaults and maximum values:

Default and maximum lifetime (days)	100
Maximum extensions	99

Examples:

Command	Action
ws_allocate my_workspace 100	Allocate a workspace named "my_workspace" for 100 days.
ws_list	List all your workspaces.
ws_find my_workspace	Get absolute path of workspace "my_workspace".
ws_extend my_workspace 100	Extend life me of workspace my_workspace by 100 days from now.
ws_release my_workspace	Manually erase your workspace "my_workspace" and release used space on storage (remove data first for immediate deletion of the data).

Sharing Workspace Data within your Workgroup

Data in workspaces can be shared with colleagues. Making workspaces world readable/writable using standard unix access rights is strongly discouraged. It is recommended to use ACL (Access Control Lists).

Best practices with respect to ACL usage:

• Take into account that ACL take precedence over standard unix access rights
• Use a single set of rules at the level of a workspace
• Make the entire workspace either readonly or readwrite for individual co-workers
• Optional: Make the entire workspace readonly for your Rechenvorhaben (group bwYYMNNN), e.g. for large input data
• If a more granular set of rules is necessary, consider using additional workspaces
• The owner of a workspace is responsible for its content and management

Examples:

Command	Action
getfacl $(ws_find my_workspace)	List access rights on the workspace named "my_workspace"
setfacl -Rm u:fr_xy1001:rX,d:u:fr_xy1001:rX $(ws_find my_workspace)	Grant user "fr_xy1001" read-only access to the workspace named "my_workspace"
setfacl -Rm u:fr_xy1001:rwX,d:u:fr_xy1001:rwX $(ws_find my_workspace)	Grant user "fr_xy1001" read and write access to the workspace named "my_workspace"
setfacl -Rm g:bw16e001:rX,d:g:bw16e001:rX $(ws_find my_workspace)	Grant group (Rechenvorhaben) "bw16e001" read-only access to the workspace named "my_workspace"
setfacl -Rb $(ws_find my_workspace)	Remove all ACL rights. Standard Unix access rights apply again.

Local Disk Space

All compute nodes are equipped with a local SSD with 240 GB capacity (usable 200 GB). During computation the environment variable $TMPDIR points to this local disk space. The data will become unavailable as soon as the job has finished.

High Performance Network

The compute nodes all are interconnected through the high performance network Omni-Path which offers a very small latency and 100 Gbit/s throughput. The parallel storage for the work spaces is attached via Omni-Path to all cluster nodes. For non-blocking communication 17 islands with 44 nodes and 880 cores each are available. The islands are connected with a blocking factor of 1:11 (or 400 Gbit/s for 44 nodes).