Main Navigation

Secondary Navigation

Page Contents

Contents

Resources of the High Performance Computer 'Elwetritsch'

Hardware Complexes

During time different hardware entered the High Performance Computer. To guarantee reproducibility and for parallel job a homogeneous environment, these different hardware components are clustered in complexes. Special among the complexes are head and dedicated which are reserved and not usable for longer jobs. The head nodes are intended for interactive work and dedicated nodes are owned by others and only available for their owners.

Complex Server Properties Summary
Host Type CPUs Cores Memory Interconnect # Servers ∑ Cores ∑ Memory
head elwe1: Intel E5-2670 2 16 128 GB 10 GE + IB QDR 1 16 128 GB
elwe2: Intel E5-2670 + 2x NVIDIA Tesla M2090 2 16 128 GB 10 GE + IB QDR 1 16 128 GB
elwe3: Intel E5-2670 2 16 128 GB 10 GE + IB QDR 1 16 128 GB
elwe4: Intel E5-2640v3 + 1x NVIDIA Tesla K80x 2 16 256 GB 10 GE + IB QDR 1 16 256 GB
4 64 640 GB
Complex Server Properties Summary
Host Type CPUs Cores Memory Interconnect # Servers ∑ Cores ∑ Memory
mpi32 Intel E5-2670 2 16 32 GB IB QDR 158 2528 5056 GB
mpi64 Intel E5-2670 2 16 64 GB IB QDR 116 1856 7424 GB
mpi128 Intel E5-2670 2 16 128 GB IB QDR 18 288 2304 GB
mpi64v3 Intel E5-2640v3 2 16 64 GB IB QDR 96 1536 6144 GB
mpi256v3 Intel E5-2640v3 2 16 256 GB IB QDR 24 384 6144 GB
single64v3 Intel E5-2637v3 2 8 64 GB IB QDR 24 192 1536 GB
accelerator Intel E5-2670 + 2x NVIDIA Tesla M2090 2 16 32 GB IB QDR 3 48 96 GB
Intel E5-2670 + 2x NVIDIA Tesla M2090 2 16 64 GB IB QDR 3 48 192 GB
Intel E5-2670 + 2x NVIDIA Tesla K20x 2 16 64 GB IB QDR 2 32 128 GB
Intel E5-2670 + 4x NVIDIA Tesla K20x 2 16 64 GB IB QDR 1 16 64 GB
Intel E5-2670 + 2x Intel XEON Phi 5110P 2 16 32 GB IB QDR 1 16 32 GB
Intel E5-2670 + 2x Intel XEON Phi 5110P 2 16 64 GB IB QDR 2 32 128 GB
bigmem Intel XEON E5-4650 4 32 1024 GB IB QDR 1 32 1024 GB
vgl Intel E3-1270 + NVIDIA Quadro 4000 1 4 16 GB IB QDR 9 36 144 GB
dedicated 36 600 3576 GB
494 7644 34088 GB

Queues

Jobs are organized in queues or partitions according to their resource requirements. Best practice is not to use queues or partitions for job submission but resource requirements and leave the selection of a queue to the batch system. A very important resource is the job run time. It should be specified approximately, see guide. Other important resources are the requested memory of a job or the hardware of a CPU, all described in the guide.

File Systems

The High Performance Computer comes up with different file systems for different purposes and policies. Please be aware that all files are deleted after your RHRK account has expired or your project has ended.

Name Tier Priority State Nodes Job Slots Max Walltime Running Suspended Pending
PathTypeSizeFreeCluster-wideQuotaBackupPurpose
$HOMENFS9 TB6 TByes40 GBvia NetApp snapshots
in /home/.snapshot
personal files, self-compiled programs
(will be deleted when your RHRK account expires)
$SCRATCHFhGFS321 TB49 TByesnonenolarge files that can be read and written in parallel
(oldest files will be deleted when file system is 85% full)
$WORKNFS20 TB6 TByesnonenosmall files for simulations - on request only
(will be deleted when your RHRK account expires)
$TMPDIRext4varyingn/anononenoshort-time storage of local files
(will be deleted after job ends)