Measuring HPC: Performance, Cost, & Value

Experts in numerical software and
High Performance Computing
Measuring HPC:
Performance, Cost & Value
IDC HPC User Forum
Austin, TX
6-8 September 2016
Andrew Jones
Vice-President Strategic HPC Services

3High Performance Computing Consulting | Numerical Algorithms | Software Engineering Services | www.nag.com
Instead, use your time to
optimize benefit from those decisions make sure the next decision is right
Weak idea
Don’t waste your valuable time justifying past decisions
“Justify the investments made”

Optimize usage to maximize
science/engineering impact and business value
Where possible, “fairness” innovation vs. production?
Understand how the current HPC capability is used

Use data on value delivered by
current HPC capability
so future investments can be
made with confidence
at the right scale, type and
timing for maximum value

Make best use of current capability
Make best decisions for the future capability

Benchmarks
“How fast is my
code?”
“How fast is this
supercomputer?”
“Which
system/processor
is faster?”
“How does my
code scale?”

Drive value for money + technical confidence
Guide
procurement
planning
Support
procurement
evaluation
Acceptance
testing

0
2
4
6
8
10
12
14
A B C D E F G
Performance
0
2
4
6
8
10
12
14
A B C D E F G
Performance

How well is the HPC facility/service run?
How well is it used?
How fair is it?
Cost, business impact, … [in next section]

How effectively do teams work together?
Productivity and happiness of users, service staff, etc.
Organizational structures
Governance
Etc.

… but also affects everyone
Performance becomes a discipline in its own right …
Lots of things to measure!
Not just code speed

Computer Storage Network Staff Electricity
Cooling Facility costs Security Time (!) Resilience
Measurement &
Reporting
Financing effects Insurance
Irrecoverable
taxes
Procurement
project
Application
porting
Software licenses
User training &
support
Documentation Commissioning
Warranty,
support,
maintenance
Decommissioning
Availability /
throughput
Data migration … and more …

?
Which costs
can you
control?
Which costs
does your
budget bear?
How are you
incentivized
against other
costs?
How will costs
change over
time?

Total Cost of Ownership
(TCO)
• What are all the costs
involved in owning and
operating this HPC service?
• Most important to HPC
owners/managers
• It costs money even if not
used (always)
Total Cost of Solution
(TCS)
• What are all the costs I will
incur using this HPC service
for a specific project?
• Most important to HPC
users
• It costs nothing if not used
(sometimes)

Much more complicated than just price of hardware …
Which costs can be controlled?
TCO vs. TCS

ROI is only part of the answer
Value to the business might be binary:
project/product not possible without HPC
Determining value can be hard & sometimes subjective

Measuring HPC can be very powerful
for the business and the user when done well
Measuring performance, costs and value
are all complex and can be hard to get right
Measurement is only the start …
effective reporting can be even more important!
NAG HPC consulting can help you

Experts in High Performance Computing,
Algorithms and Numerical Software Engineering
www.nag.com | blog.nag.com | @NAGtalk
@hpcnotes
Please come to our tutorial at SC16 in Austin this November:
Essential HPC Finance Practice – TCO, cost recovery models, etc.

Measuring HPC: Performance, Cost, & Value

More Related Content

Viewers also liked (20)

Similar to Measuring HPC: Performance, Cost, & Value (20)

More from inside-BigData.com (20)

Recently uploaded (20)

Measuring HPC: Performance, Cost, & Value