This document summarizes the benefits of High Performance Computing (HPC) and Dell's HPC solutions. It states that HPC is critical for solving big problems that require faster answers through integrated and scalable solutions. It then describes Dell's end-to-end HPC services including modular designs, expertise, and tools. Case studies show how Dell HPC solutions have benefited manufacturing, healthcare, finance, and energy customers. The document concludes by stating that Dell aims to help more organizations use HPC to drive innovation.

1. MT10: Give Your organization better, faster
insights and answers with High Performance
Computing (HPC)

2. HPC is imperative to research, industry and security.
“HPC systems… are vital to the
Nation’s interests in science,
medicine, engineering, technology,
and industry. The NSCI will spur the
creation and deployment of
computing technology at the
leading edge, helping to advance …
economic competiveness,
scientific discovery, and national
security.” (July 29, 2015)
Now a U.S. top priority:
National Strategic
Computing Initiative (NSCI)

3. HPC is critical for BIG problems that require faster,
better answers via integrated, scalable solutions.
Cloud Computing
Emerging usage model to
simplify HPC
Modeling and
simulation
Traditional Parallel
Computing Clusters
Large ensemble problems
Analyze multiple data sets
High throughput
computing
Fast answers and insights for
data intensive problems
Big data
analytics

4. Scalable, end-to-end approach that enables faster insights
and competitive advantage
Lifecycle
services
Modular and
reference
designs
Flexible
architectures
HPC domain
expertise
• Faster time to discovery
End to end services maximize your HPC investment, from financing
to configuration, installation, cluster management and recycling.
• Integrated systems
Engineered and packaged solutions are fine-tuned for use cases.
• Standards-based building blocks
Modular architecture grows capacity as needed.
• 31 years of computer industry experience
Dell is a proven HPC partner, offering standard, open solutions from
Dell and strategic partners.
• Optimized portfolio of industry leading tools
Augments and adapts to customers’ operations.
• Dell HPC engineering and research team
Focus on your business goals.
• Best practices, performance characterizations
Incorporate Dell research into each HPC installation.

5. Select, deploy, manage and scale your cluster with a single source of support.
Get exactly the help you need.
Rack Integration
Services
Dell HPC Merge Center
15 of the world’s top 500
reported supercomputers
were built at this facility in
Austin. Five were
interconnected onsite.
• Expand IT capabilities to
scale with your business
• Speed up deployments.
and improve
productivity.
• Optimize performance
with flexible, customized
services.
Turnkey outsource system
management service that
provides secure remote
system monitoring,
administration and support
to help increase your
system utilization and
uptime.
Dell Financial Services
provides flexible payment
options so you can get
more compute power for
your money. By providing
the financing, we make it
easier for you to upgrade
and refresh.
Deployment
Services
Remote Cluster
Management Services
Dell Financial
Services

6. We lead and innovate in HPC, so you can focus
on discoveries and impact
Dell will help more people in research, industry and
government use HPC solutions to make more innovations
and discoveries that improve society and advance human
understanding than any other HPC systems vendor in the
world.

7. HPC use case:
manufacturing
Challenges
• Improve engineering productivity.
• Reduce design time and costs.
• Enable new breakthrough research.
• Manage third party engineering applications
for a global workforce.
Results
The combination of a well designed, balanced
system with supported software and services is
allowing manufacturer to deliver ground-breaking
HPC resources to engineering community.

8. HPC use case:
healthcare/life sciences
Challenge
• Achieve remission in “incurable” pediatric
cancer patients by developing customized
treatments fast enough to help patients with
very limited intervention time frames.
Results
• Reduce genomic sequencing time from weeks
to days.
• Deliver individualized therapies in less than a
week.
“We’ve gone from treating every child exactly the same way
to being able to develop individualized therapies. We’re now
able to stop the progression of cancer in 60 percent of our
patients, and today some are cancer-free.”
— Dr. Giselle Sholler, Chair of the NMTRC and Director of the Pediatric
Oncology Research Program at Helen DeVos Children’s Hospital

9. HPC use case:
financial services
Challenge
• Speed up financial analyses and stock
market trading analyses that use floating
point applications.
• Process huge volumes of transactional
data on a daily basis.
Results
• Powerful cluster based on PowerEdge C6220
servers processes runs all transactional data
and real time interest rate information.
Processor speed is critical.
• Business is in process of cluster expansion by
adding additional GPU accelerator capabilities
to speed up modeling and Monte Carlo
simulations.

10. HPC use case:
energy/oil and gas
Challenge
• Locate profitable new deep water oil and gas
fields globally.
Results
GPU-based HPC system enables oil and gas
companies to make intelligent decisions prior to
beginning costly deep sea projects.
• CPU processing is handled via
Schlumberger Omega software with many
different algorithms.
• GPU processing is used for specific
algorithms such as Kirchhoff Depth
Migration, Reverse Time Migration,

11. Dedicated to research and development of HPC solutions
Dell HPC Solutions Engineering Lab
Meet real-life,
workload-specific
challenges
through
collaboration with
the global HPC
research
community.
• Dell XL Consortium
• Dell | Cambridge HPC
Solution Centre
• National Center for
Supercomputing
Applications (NCSA) —
University of Illinois Private
Sector Program
• San Diego Supercomputing
Center (SDSC)
• University of Pisa
• University of Texas — Texas
Accelerated Computing
Center (TACC)

12. Gateways to Discovery:
Cyberinfrastructure for the
Long Tail of Science
Shawn Strande
Project Manager & Co-PI Deputy Director, SDSC

13. Overview of talk
1. Motivation
2. Key features
3. Architecture
4. Highlights of research results
“Comet is all about providing
high performance
computing to a much larger
research community – what
we call ‘HPC for the 99
percent’ – and serving as a
gateway to discovery.”
– Mike Norman,
SDSC Director
1. Motivation
2. Key features
3. Architecture
4. Highlights of
research
results

14. Call to serve the “Long Tail of Science”
NSF’s solicitation 13-528 “High Performance Computing
System Acquisition: Building a More Inclusive Computing
Environment for Science and Engineering”
• Expand the use of high end resources to a
much larger and more diverse community.
• Support the entire spectrum of NSF
communities.
• Promote a more comprehensive and
balanced portfolio.
• Include research communities that are not
users of traditional HPC systems.

15. Data extracted from NSF’s XDMoD database informed a design that
reflects the way researchers actually use HPC systems.
• 99% of jobs run on
NSF’s HPC resources
in 2012 used <2,048
cores.
• Consumed >50% of
the total core-hours
across NSF resources.
One rack of Comet
A system designed to serve the 99% is
significantly different than one for the 1%
99% 1%

16. Comet’s integrated architecture is a platform for
a wide range of computing modalities.
128 GB/node, 24 core
nodes support shared
jobs & reduce the need
for runs across racks.
99% of the jobs run
inside a single rack with
full bisection BW.
High performance and
durable storage support
compute & data
workflows, with replication
for critical data
Support science
gateways as a
primary use case.
Compute, GPU, and
large memory nodes
support diverse
computing needs.
Virtual Clusters give
communities control
over their software
environment.

17. InfiniBand compute, ethernet storage
Comet network architecture
Juniper
100 Gbps
Arista
40GbE
(2x)
Data Mover
Nodes
Research and
Education Network
Access
Data Movers
Internet 2
7x 36-port FDR in each rack
wired as full fat-tree. 4:1 over
subscription between racks.
72 HSWL320 GB
Core
InfiniBand
(2 x 108-
port)
36 GPU
4 Large-
Memory
IB-Ethernet
Bridges (4 x
18-port each)
Performance Storage
7.7 PB, 200 GB/s
32 storage servers
Durable Storage
6 PB, 100 GB/s
64 storage servers
Arista
40GbE
(2x)
27 racks
FDR 36p
FDR 36p
64 128
18
72 HSWL320 GB
72 HSWL
2*36
4*18
Mid-tier
InfiniBand
Additional support ccomponents
(not shown for clarity)
Ethernet management network (10 GbE)
Node-Local
Storage 18
72FDR
FDR
40GbE
40GbE
10GbE
18
switches
4
4
FDR
72
Home File Systems
VM Image Repository
Login
Data Mover
Management
Gateway
Hosts

18. Comet is built for High Performance Computing.
27 Rack-sized building blocks
• Each Rack: 1728 Cores, 9.2TB Memory,
23TB Flash
• Full System: 46,656 Cores, 248TB Memory,
620TB Flash
Modest heterogeneity for diverse
workflows
• 36 GPU nodes with dual NVIDIA K80s
(4 GPUs/node)
• 4 large-memory nodes: 1.5TB Memory,
64 cores
Large Ethernet connectivity to
storage and outside world
• 72 x 40GbE = 2.8 Tbit/s
• 100Gbit/s to CENIC/Internet2
Large, High-Speed Parallel I/O System
• 7.6 PB of Lustre-based Hard Disk
• > 200GB/sec, sustained

19. 27 rack-based supercomputers ~ 2.0PF/s
18 X
7 X
~ 2 X
4 Nodes/2U: 8X Haswell 12Core 2.5GHz,
0.5TB Memory
Mellanox FDR (56Gb/s) InfiniBand
48 Port 10GbE + 2 x 40GbE
• Single rack supports > 95% of
XSEDE applications.
~75 TF/s / Rack
• Full bisectional cluster
interconnect (InfiniBand)
18 uplinks/Rack (4:1
oversubscription)
• 10GbE management network

20. Ethernet connectivity – 2.8 Tbit/s
2 Arista 7508 Enterprise
Ethernet Switch-Routers
72 x 40GbE
CometNodesviaIB
72 x FDR IB
Mellanox IB  Ethernet
64 x
40GbE
7.6 PB, 200GB/sec Parallel File
System
100GbE
Connections to Scientists

21. Virtualized HPC — user-customized HPC
Frontend
Virtual Frontend
Hosting Disk Image Vault
Compute
Compute
Compute
Compute
Compute
Compute
Compute
Compute
Compute
public network
private
Virtual
Frontend
Virtual Compute
Virtual Compute
Virtual Compute
private
Virtual
Frontend
Virtual Compute
Virtual Compute
Virtual Compute
private
physical
virtual
virtual

22. High performance virtual cluster characteristics
Virtual
Frontend
Virtual
Compute
Virtual
Compute
Virtual
Compute
private
Ethernet
InfiniBand
All nodes have
• Private Ethernet
• InfiniBand
• Local Disk Storage
Virtual Compute Nodes can network boot (PXE)
from its virtual front end.
All Disks retain state
• Keep user configuration between boots.
InfiniBand virtualization
• 8% latency overhead
• Nominal bandwidth overhead
Comet – Pathfinder for
Virtualized HPC in XSEDE

23. Comet serving science and society
DNA nanostructures
Seismic research and
disaster prevention
DNA nanostructures
CIPRES — Assembling
the Tree of Life
Social sciences
Neurosciences and
brain research
Astrophysics
Fluid Turbulent
Physics
Alternative energy
New Materials
Research
Climate change and
environmental
sciences
Molecular science

24. Comet’s operational polices and software are
designed to support long tail users.
• Optimized for throughput
• Science gateways reach large communities.
• Virtual clusters (VC) will support well-formed communities

25. Highlights of research using
Comet

26. Simulations of biological membranes
Wonpil Im (U. Kansas) has
been making extensive use of
Comet to perform molecular
dynamics simulations on
biological membranes to
study their mechanical
properties and the
interactions between lipids
and proteins.
O
OH
HO
O
OH
O
OH
HO
OH
OH
O
HO
OH
O
O
O
O
NHAc
COO-
HO
OH
OH
OH
O
O
AcHN
OH
OH
Gal1
Neu5Ac
Gal2
GalNAc
Glc
CER
(A) (B)
This work can lead to a better understanding of how amyloid plaques
form in the brains of Alzheimer patients.

27. 2D compressible turbulence
Alexei Kritsuk (UCSD) has been using Comet to study compressible
turbulence in two dimensions.
This research can provide insights into the structure of the universe.
(Kritsuk and Falkovich, 15th European Turbulence Conf. 2015)

28. Colloids and self-assembling systems
Sharon Glotzer (University of
Michigan) uses Comet to simulate
colloids of hard particles, including
spheres, spheres cut by planes,
ellipsoids, convex polyhedra,
convex spheropolyhedra, and
general polyhedra.
Glotzer’s work can lead to the design of better materials, including
surfactants. liquid crystals and nanoparticles that spontaneously assemble
into sheets, tubes, wires or other geometries.

29. Protein lyophilization (freeze-drying)
Pablo Debenedetti (Princeton)
uses Comet to study
lyophilization (freeze-drying), a
standard technique used to
increase the storage life of
labile biochemical, including
therapeutic proteins, by the
pharmaceutical industry.
Top left: Trp-cage miniprotein structure. Top right: Mean-squared fluctuation for each residue in
Trp-cage for the hydrated and dehydrated powder system. Bottom left: Lysozyme protein
structure. Bottom right: Water sorption isotherm for lysozyme.

30. Studying flu at the molecular scale
Rommie Amaro (UCSD) uses Comet to understand how properties of the
flu virus affect their infectivity.
Alasdair Steven, NIH
Brownian dynamics illuminates how glycoprotein stalk height impacts
substrate binding.

31. Comet is a partnership of
academia, the National Science
Foundation, industry and the
XSEDE user community

32. Users via gateways outnumber those logging in.
Many other
gateways are taking
off with hundreds
of thousands of
users
(nanoHUB, Galaxy,
Folding@Home,
more).

33. SEAGrid applications & usage on SDSC Comet, the workhorse
Vortex shedding with Nek5000
Jeilani,Y et al Phys. Chem.
Chem. Phys. 2015
Courtesy Sudhakar Pamidighantam
Badieyan, S et al. J. Org. Chem. J. Org. Chem., 2015, 80 (4),
PI Sudhakar Pamidighantam, Indiana U,
www.seagrid.org
611 Users, 24000 Jobs, 11.5M XSEDE SUs used since April 2014
More than 120 Publications, 13 Dissertations

34. Comet is delivering on its promise.
• Users are benefiting from Comet’s rapid
turnaround and performance.
• Comet is becoming the
workhorse system in XSEDE
for science gateways.
• Our partnership with
Indiana University will lead
to high performance
virtual clusters.
“I applaud your recent launch of the "Comet" platform
which acknowledges what most scientific computing
really looks like…” — Carl Boettiger, UCB
“In general, Comet has become one of the most reliable and
productive clusters that the Ultrascan gateway uses.” — Gary
Gorbet, University of Texas Health Science Center
“..have you heard about the 1000 CPU-hour allocations? They’re great!...I
never knew how long the queues were because every time I looked at my
job it was running, Comet is a great machine.”
—Ted Wetherbee, Fond du Lac Tribal and Community College

35. Closing thoughts
Research organizations
must move faster and
better than ever before.
Integrated, cost-effective, flexible
solutions that scale are critical to
solving BIG problems that involve:
• Complex modeling and
simulation.
• Big data analytics.
• Large ensemble problems and
analyzing multiple data sets.
• Emerging usage models that
simplify HPC access.

36. Take your next steps
to accelerate your
high performance
computing. Start today
Schedule time with a Dell HPC
Solutions Specialist
For more information, visit us online:
• Dell.com/hpc
• HPCatDell.com

37. Thanks!