Bee cluster

Overview

Bee is a university cluster available to ICP-affiliated personnel. Software can be loaded via Spack.

All of the storage available on the cluster is handled with a BeeGFS file system. In total, we provide 144 TB of storage shared by all users. Note that each user is not given a specified amount of storage on the cluster and this can cause issues when it is expended. See cluster status page for real-time stats.

In total, the cluster consists of 768 compute cores split across 45 nodes using a 40 Gbps InfiniBand.

Type

Nodes

Cores

RAM

GPU

Node list

Standard

42

2x8

128 GB

1x Titan Black

compute-[x]-[y]

Big

2

4x10

512 GB

compute-big-[x]-[y]

GPU

1

2x8

128 GB

4x Tesla K20C

compute-quadgpu-[x]-[y]

This cluster is managed by the ICP. If there are any issues, please submit an issue to the ICP cluster team. The cluster was acquired via a DFG grant in 2014 [1].

Partitions and nodes

Contrary to most clusters, compute resources on Bee are not segmented in partitions. Instead, one has to select resources from a node list.

Standard nodes

There are 42 standard nodes equipped with 2 Intel Dual Xeon E5-2630v3 (8 cores, 18 threads, 2.4 GHz, 85 W), 128 GB of RAM, InfiniBand (40 Gbps) interconnect. Each node has 1 NVIDIA Titan Black (6 GB GDDR5). All standard nodes follow the naming convention compute-[x]-[y] where they can take on values compute-0-1 ... compute-0-17 and compute-1-0 ... compute-1-23.

Big nodes

There are 2 big nodes equipped with 2 Intel Quad Xeon E5-4640v2 (10 cores, 20 threads, 2.2 GHz, 95 W), 512 GB of RAM, InfiniBand (40 Gbps) interconnect. There are no GPUs. All big nodes follow the naming convention compute-big-[x]-[y] where they may take on values compute-big-0-0 and compute-big-0-1.

GPU node

There is 1 quad-GPU node equipped with 2 Intel Dual Xeon E5-2630v3 (8 cores, 18 threads, 2.4 GHz, 85 W), 128 GB of RAM, InfiniBand (40 Gbps) interconnect. The node has 4 NVIDIA Tesla K20C (5 GB GDDR5). The node name is compute-quadgpu-1-0.

Access

Frank Huber is in charge of creating user accounts on Bee. All ICP students and staff members are entitled to accessing Bee. For external personnel, Christian Holm must personally approve access. To log into Bee, refer to Logging into clusters. See also Using Bee for building software and submitting jobs.

Obligations

The only obligation is to acknowledge funding from the DFG. Since the project was awarded before 2017, there is no obligation to cite the grant id. However, if you would like the publication to appear in grant search results, here is the complete acknowledgment:

We acknowledge funding from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) through the Compute Cluster grant no. 261833929.

Publications