HITS – computational sciences on the way into the cloud?aws-de-media.s3.amazonaws.com/images/AWS...
Transcript of HITS – computational sciences on the way into the cloud?aws-de-media.s3.amazonaws.com/images/AWS...
HITS – computational sciences on the way into the cloud? Frauke Gräter HITS Heidelberg
Heidelberg Institute for Theoretical Studies
HITS
Heidelberg Institute for Theoretical Studies
Klaus Tschira Stiftung: Supports informatics, natural sciences, mathematics
Klaus Tschira 1972 - Co-founded the German software giant SAP AG 1995 – Bought Villa Bosch, founded Klaus Tschira Stiftung
Setup: ● 110 scientists in 12 research groups ● non-profit ● from physics to biology to computer science ● strong ties in Heidelberg and world-wide Agenda: multidisciplinary computational sciences ● Molecular modeling and simulation ● Computational biology ● Bioinformatics & databases ● Computational linguistics ● Theoretical astrophysics ● Computational Statistics, big data
Heidelberg Institute for Theoretical Studies
big data big scales big challenges
Compute Infrastructure @ HITS
• 110 scientists in 12 research groups • Small HPC cluster for testing:
– 5732 cores – 5 nodes with nVidia Kepler GPUs,
2 nodes with Intel Xeon Phi – 28TB RAM, one node with 4TB RAM – QDR Infiniband – 1.8PB storage
à No scale out possibility due to rack space
à Main HPC workloads done in (inter)national supercomputing centers
Molecular Biomechanics group, Frauke Gräter
Forces in biology!
blood spider web muscle focal adhesion
Proteins in the crash test
macro (meter) scale
molecular (nano) scale
Mechanics of large structures
The challenge:
The challenge:
. . . . .
Gromacs (or NAMD): scale up to ~500 atoms / processor
typical demand:
50-1,000 CPUs (& GPUs) TBs data
HITS in the cloud? MD simulations
cores freq performance normalized
HITS 24 cores, 48 HT 1.8Ghz 47ns/day 47ns/day AWS c4.8xlarge 36vCPU 2.9GHz 60ns/day 49ns/day
41,000 atoms, Gromacs 5.0 AVX2 optimized binaries
HITS in the cloud? MD simulations
SMALL: shared HITS internal cluster: ~3,000 cores LARGE: supercomputing infrastructure (Stuttgart, Munich, PRACE): ~2-20 Mio CPU hours / project
SMALL-SIZED but MANY! Gromacs on AWS: flexible c4 instances: #CPUs = #hyperthreaded cores
com
puta
tiona
l cos
t
Current AWS developments @ HITS
• Covering peak requests on development cluster with EC2 instances
• Archiving of results on Glacier
• Current developments: – Transparent scheduling of independent jobs on hybrid
computing environment using spot market instances – Budget/Quota-management within research teams – Collaborative working platform for large datasets with
international partners, e.g. Square-Kilometre-Array (SKA) – Genome assembly with huge temporal storage requirements
(PB-scale) – Molecular Dynamics simulations, mid-sized instances
13
More information: www.h-its.org Twitter: @HITStudies Facebook: /HITStudies Youtube: /TheHITSters