T. Fisher - HPC Cloud: Hype or Reality

Technical Computing
HPC Cloud: Hype or Reality?
Terry Fisher
Global Sales Leader - HPC Cloud
IBM Platform Computing
CHPC Conference –
December 2013
© 2013 IBM Corporation
Technical Computing
Based on HPC cloud’s potential impact, many organizations are
evolving their infrastructures to enable private cloud deployments,
exploring hybrid clouds, and considering public clouds
Evolve existing
infrastructure to
HPC cloud to enhance
flexibility, and
cost effectiveness.
Enable integrated
approach to improve
HPC cost and
Access additional
HPC capacity with
variable cost model
© 2013 IBM Corporation
Technical Computing
HPC clouds may need to address advanced requirements that
extend beyond core HPC workload management on a cluster or grid
If you have ever wondered:
• Can I share infrastructure between multiple clusters (simulation, design,
analytics, Big Data) in one integrated environment?
• Can I move resources (compute nodes) from one cluster to another when the
clusters are already deployed?
• Can I selectively run specific HPC jobs in dynamically created VMs?
• Can I re-provision my infrastructure based on my workload requirements?
• Can I extent my HPC environment by bursting to a public cloud for peak loads?
• Can I support a new business model and deliver an analytics or HPC service in
a multi-tenant environment?
With IBM Platform Computing solutions, the answer to all of these
questions is “YES”!
© 2013 IBM Corporation
Technical Computing
What requirements does a HPC cloud fulfill?
• Provides additional compute resources to meet peak demand (bursting)
• Dynamically resizes HPC cluster based on demand (flexing)
• Allows technical end users to run HPC jobs on-demand
• Enables multiple groups to share an HPC environment managed by business
policies with chargeback and reporting
• Dynamically changes HPC cluster configurations (e.g. OS personality) based on
workload requirements
• Enables end users to self-service create custom HPC clusters on-demand
Fulfilling NIST Cloud Characteristics of enabling convenient, on-demand network access to
a shared pool of configurable computing resources that can be rapidly provisioned and
released with minimal management effort or service provider interaction
Note: HPC cloud does not necessarily imply a public cloud deployment model or VMs
© 2013 IBM Corporation
Technical Computing
HPC Cloud Customer Examples
© 2013 IBM Corporation
Technical Computing
HPC Cloud - Computer Aided Engineering (CAE) Customer Example
Customer Background:
• A developer and distributor of Computer
Aided Engineering software with a focus on
computation fluid dynamics
• Primary I/T environment run on IBM servers
with GPFS in an IBM ITS data center
• Preparing to offer a new SaaS option
Customer Goals Addressed:
• Enhancing “foundational” I/T capacity (e.g. servers, storage) for internal and external use
with high-capacity
• Establishing portal capability for self-service
• Enabling multi-tenancy with remote visualization
• Future extension to external “variable” I/T capacity for an integrated ability to support peak
© 2013 IBM Corporation
Technical Computing
HPC Cloud – Higher Education Customer Example
Customer Background:
• Resources for technical calculations are distributed in various
university institutes
• Distributed workstations and compute servers are unevenly utilized
• Faculty and researchers are unable to effectively perform both
calculation intensive workloads (e.g. Ansys) and interactive work (e.g.
Matlab) on their existing systems
• Current I/T systems are heterogeneous
• Which software is on which machines is not catalogued
Customer Goals Addressed:
• An automated system providing technical computing resources to
authorized users quickly and with little effort.
• User self-service available through a GUI portal
• Dynamic provisioning of physical and virtual machines
• Supporting complex simulations and data intensive analysis
• Automated instantiation of needed software
• Ability to extend in future to access external resources
© 2013 IBM Corporation
Technical Computing
HPC Cloud – Petroleum Customer Use Case Example
Customer Background:
• Seismic processing required to make multimillion dollar decisions on
where to drill or how to best exploit oil fields
• Seismic processing requirements for large amounts of processing power
• Need for extensive data cleansing to filter out the noise measured along
with the signals received by detectors
• The ability to handle and store large volumes of data
Customer Goals Addressed:
• On demand access to a cloud infrastructure to process very large
amounts of seismic data quickly
• Access to advanced HPC bare metal servers with the latest advanced
high-core count Intel CPUs
• Leverage of GPU nodes to accelerate workload processing up to 10X
• Scalability up from a base of 120 nodes and back down again as
needed all within a single location to avoid network bottlenecks
© 2013 IBM Corporation
Technical Computing
HPC Hybrid Cloud Enablement
© 2013 IBM Corporation
Technical Computing
Representative HPC Challenges Addressed
Lack of compute capacity
during peak demand
Lack of agility is impacting
the business
Unsteady workload
How to deal with workload
Capital expense to address
peak demand is too high
Lack of Disaster Recovery
© 2013 IBM Corporation
Technical Computing
HPC SaaS Client Use Cases
• Cloud Burst target for existing clients
– The IBM Platform HPC SaaS offering will provide a cloud based burst target for
existing Platform LSF and Symphony clients to extend or grow their on-premise
– When additional HPC capacity is needed, on-premise cluster resources will connect to
a cluster running in the cloud
– Additional capacity is desired when the on-premise cluster is temporarily overloaded
with too many pending jobs, special projects are scheduled that will exceed the
current capacity, and when access to special hardware not available on the onpremise cluster is needed (e.g., GPU, large memory, etc)
© 2013 IBM Corporation
Technical Computing
Administrator and Calendar Driven Bursting to HPC Cloud
User A
User B
User C
User N
On-premise LSF or Symphony cluster
Portal to launch
Cluster on Demand
Administrator and
Calendar Driven
LSF or
Temporary Platform LSF or
Platform Symphony Cluster
SoftLayer Cloud
Bare-Metal, Private, Public
Note: Planned availability 1Q14
IBM Confidential
© 2013 IBM Corporation
Technical Computing
Policy Driven Bursting to HPC Cloud
User A
User B
User C
User N
On-premise LSF or Symphony cluster
Grid Load
LSF or
Temporary Platform LSF or
Platform Symphony Cluster
SoftLayer Cloud
Bare-Metal, Private, Public
Note: Planned availability 2Q14
IBM Confidential
© 2013 IBM Corporation
Technical Computing
Platform Computing SaaS Client Use Cases
• IBM Platform SaaS Offering
– Fully functioning and ready to use Platform LSF and Platform Symphony clusters that
will be installed and operated in the Cloud (i.e. SoftLayer data centers)
– Cloud bare metal machine instances, the operating systems, and the cluster software
- Platform LSF or Platform Symphony.
– Use as needed and pay for only the time they use the cluster, instead of purchasing
perpetual licenses
– Rent the cluster on a pay per use basis as an OPEX and not need to make a CAPEX
investment in hardware and software
• Hosted HPC for new clients
– Provides an HPC (Platform LSF or Platform Symphony) cluster to new clients who do
not have an on-premise HPC cluster, or the skills and budget to build one
– When access to a HPC cluster is required, access a cluster running in the cloud to
process workload
– Provide an Analytics cluster (Platform Symphony, BigInsights) for new clients who
need to perform Hadoop/MapReduce work
© 2013 IBM Corporation
Technical Computing
IBM Platform Computing enables HPC Cloud business and IT value
© 2013 IBM Corporation
Technical Computing
Wrap-up Questions?
© 2013 IBM Corporation
Technical Computing
Visit us at:
© 2013 IBM Corporation

similar documents